agora inbox for [email protected]
help / color / mirror / Atom feedRe: Command Triggers, patch v11
9629+ messages / 6 participants
[nested] [flat]
* Re: Command Triggers, patch v11
@ 2012-03-13 11:22 Andres Freund <[email protected]>
2012-03-13 15:14 ` Re: Command Triggers, patch v11 Alvaro Herrera <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
0 siblings, 2 replies; 9629+ messages in thread
From: Andres Freund @ 2012-03-13 11:22 UTC (permalink / raw)
To: pgsql-hackers; +Cc: Dimitri Fontaine <[email protected]>
Hi,
I did a short review of what I found after merging master
(b4af1c25bbc636379efc5d2ffb9d420765705b8a) to what I currently fetched from
your repo (d63df64580114de4d83cfe8eb45eb630724b8b6f).
- I still find it strange not to fire on cascading actions
- I dislike the missing locking leading to strange errors uppon concurrent
changes. But then thats just about all the rest of commands/* is handling
it...
T1:
BEGIN;
ALTER COMMAND TRIGGER cmd_before SET DISABLE;
T2:
BEGIN;
ALTER COMMAND TRIGGER cmd_before SET DISABLE;
T1:
COMMIT;
T2:
ERROR: tuple concurrently updated
- I think list_command_triggers should do a heap_lock_tuple(LockTupleShared)
on the command trigger tuple. But then again just about nothing else does :(
- ExecBeforeOrInsteadOfCommandTriggers is referenced in
exec_command_triggers_internal comments
- InitCommandContext comments are outdated
- generally comments look a bit outdated
- shouldn't the command trigger stuff for ALTER TABLE be done in inside
AlterTable instead of utility.c?
- you have repetitions of the following pattern:
void
ExecBeforeCommandTriggers(CommandContext cmd)
{
/* that will execute under command trigger memory context */
if (cmd != NULL && cmd->before != NIL)
exec_command_triggers_internal(cmd, cmd->before, "BEFORE");
/* switch back to the command Memory Context now */
MemoryContextSwitchTo(cmd->oldmctx);
}
1. Either cmd != NULL does not need to be checked or you need to check it
before the MemoryContextSwitchTo
2. the switch to cmd->oldmctx made me very wary at first because I wasn't sure
its guaranteed to be non NULL
- why is there a special CommandTriggerContext if its not reset separately?
Should it be reset? I have to say that I dislike the api around this.
- an AFTER .. ALTER AGGREATE ... SET SCHEMA has the wrong schema. Probably the
same problem exists elsewhere. Or is that as-designed? Would be inconsistent
with the way object names are handled.
- what does that mean?
+ cmd.objectname = NULL; /* composite object name */
- DropPropertyStmt seems to be an unused leftover?
Andres
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
@ 2012-03-13 15:14 ` Alvaro Herrera <[email protected]>
1 sibling, 0 replies; 9629+ messages in thread
From: Alvaro Herrera @ 2012-03-13 15:14 UTC (permalink / raw)
To: Andres Freund <[email protected]>; +Cc: pgsql-hackers; Dimitri Fontaine <[email protected]>
Excerpts from Andres Freund's message of mar mar 13 08:22:26 -0300 2012:
> - I think list_command_triggers should do a heap_lock_tuple(LockTupleShared)
> on the command trigger tuple. But then again just about nothing else does :(
If you want to do something like that, I think it's probably more
appropriate to use LockDatabaseObject than heap_lock_tuple.
--
Álvaro Herrera <[email protected]>
The PostgreSQL Company - Command Prompt, Inc.
PostgreSQL Replication, Consulting, Custom Development, 24x7 support
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
@ 2012-03-13 20:07 ` Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
1 sibling, 1 reply; 9629+ messages in thread
From: Dimitri Fontaine @ 2012-03-13 20:07 UTC (permalink / raw)
To: Andres Freund <[email protected]>; +Cc: pgsql-hackers; Dimitri Fontaine <[email protected]>
Hi,
Andres Freund <[email protected]> writes:
> I did a short review of what I found after merging master
Thanks!
> - I still find it strange not to fire on cascading actions
We don't build statement for cascading so we don't fire command
triggers. The user view is that there was no drop command on the sub
objects, only on the main one.
I know it's not ideal, but that's a limit we have to bite for 9.2
unfortunately.
> - I dislike the missing locking leading to strange errors uppon concurrent
> changes. But then thats just about all the rest of commands/* is handling
> it...
>
> - I think list_command_triggers should do a heap_lock_tuple(LockTupleShared)
> on the command trigger tuple. But then again just about nothing else does :(
As you say, about nothing else does. I think that's a work for another
patch.
> - ExecBeforeOrInsteadOfCommandTriggers is referenced in
> exec_command_triggers_internal comments
> - InitCommandContext comments are outdated
> - generally comments look a bit outdated
Fixed.
> - shouldn't the command trigger stuff for ALTER TABLE be done in inside
> AlterTable instead of utility.c?
Right, done.
> - you have repetitions of the following pattern:
> void
> ExecBeforeCommandTriggers(CommandContext cmd)
> {
> /* that will execute under command trigger memory context */
> if (cmd != NULL && cmd->before != NIL)
> exec_command_triggers_internal(cmd, cmd->before, "BEFORE");
>
> /* switch back to the command Memory Context now */
> MemoryContextSwitchTo(cmd->oldmctx);
> }
>
> 1. Either cmd != NULL does not need to be checked or you need to check it
> before the MemoryContextSwitchTo
I've fixed that code.
> 2. the switch to cmd->oldmctx made me very wary at first because I wasn't sure
> its guaranteed to be non NULL
>
> - why is there a special CommandTriggerContext if its not reset separately?
> Should it be reset? I have to say that I dislike the api around this.
Some call sites need to be able to call those functions a dynamic number
of times. I could add a reset boolean parameter that would mostly be
true in all call site and false in two of them (RemoveObjects,
RemoveRelations), and add a new function to just reset the memory
context then.
Or maybe you have a better idea about the ideal API here?
> - an AFTER .. ALTER AGGREATE ... SET SCHEMA has the wrong schema. Probably the
> same problem exists elsewhere. Or is that as-designed? Would be inconsistent
> with the way object names are handled.
I'm surprised, here's an excerpt from the added regression tests:
alter function notfun(int) set schema cmd;
NOTICE: snitch: BEFORE any ALTER FUNCTION
NOTICE: snitch: BEFORE ALTER FUNCTION public notfun
NOTICE: snitch: AFTER ALTER FUNCTION cmd notfun
NOTICE: snitch: AFTER any ALTER FUNCTION
> - what does that mean?
> + cmd.objectname = NULL; /* composite object name */
User mapping and casts object names are composite, and I don't know how
to represent that in a single text structure.
> - DropPropertyStmt seems to be an unused leftover?
Seems so, cleaned out.
Regards,
--
Dimitri Fontaine
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
@ 2012-03-13 21:06 ` Andres Freund <[email protected]>
2012-03-14 03:41 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
0 siblings, 1 reply; 9629+ messages in thread
From: Andres Freund @ 2012-03-13 21:06 UTC (permalink / raw)
To: Dimitri Fontaine <[email protected]>; +Cc: pgsql-hackers
On Tuesday, March 13, 2012 09:07:32 PM Dimitri Fontaine wrote:
> Hi,
>
> Andres Freund <[email protected]> writes:
> > I did a short review of what I found after merging master
>
> Thanks!
>
> > - I still find it strange not to fire on cascading actions
>
> We don't build statement for cascading so we don't fire command
> triggers. The user view is that there was no drop command on the sub
> objects, only on the main one.
>
> I know it's not ideal, but that's a limit we have to bite for 9.2
> unfortunately.
Hm. Especially in partially replicated scenarios I think that will bite. But
then: There will be a 9.3 at some point ;)
> > - I dislike the missing locking leading to strange errors uppon
> > concurrent changes. But then thats just about all the rest of commands/*
> > is handling it...
> >
> > - I think list_command_triggers should do a
> > heap_lock_tuple(LockTupleShared)
> >
> > on the command trigger tuple. But then again just about nothing else
> > does :(
>
> As you say, about nothing else does. I think that's a work for another
> patch.
Not sure, that way the required work is getting bigger and bigger. But I can
live with that... I think the command trigger work will make better
concurrency safeness of DDL necessary.
> > 2. the switch to cmd->oldmctx made me very wary at first because I wasn't
> > sure its guaranteed to be non NULL
> >
> > - why is there a special CommandTriggerContext if its not reset
> > separately? Should it be reset? I have to say that I dislike the api
> > around this.
>
> Some call sites need to be able to call those functions a dynamic number
> of times. I could add a reset boolean parameter that would mostly be
> true in all call site and false in two of them (RemoveObjects,
> RemoveRelations), and add a new function to just reset the memory
> context then.
> Or maybe you have a better idea about the ideal API here?
I wonder if the answer is making the API more symmetric. Seems more future-
proof in combination to being cleaner.
//create a new memory context
InitCommandContext(cmd);
if(CommandFiresTriggers(cmd)){
//still in current memory context, after all not much memory should be
allocated here
cmd.foo = bar;
//switches memory context during function execution, resets it afterwards
ExecBeforeCommandTriggers(cmd);
}
if(CommandFiresTriggers(cmd)){
cmd.zap = blub;
ExecAfterCommandTriggers(cmd);
}
//drop the memory context
CleanupCommandContext(cmd);
I find the changing of memory context in CommandFires[After]Trigger + switch
back in Exec*CommandTrigger rather bad style and I don't really see the point
anyway.
> > - an AFTER .. ALTER AGGREATE ... SET SCHEMA has the wrong schema.
> > Probably the same problem exists elsewhere. Or is that as-designed?
> > Would be inconsistent with the way object names are handled.
>
> I'm surprised, here's an excerpt from the added regression tests:
>
> alter function notfun(int) set schema cmd;
> NOTICE: snitch: BEFORE any ALTER FUNCTION
> NOTICE: snitch: BEFORE ALTER FUNCTION public notfun
> NOTICE: snitch: AFTER ALTER FUNCTION cmd notfun
> NOTICE: snitch: AFTER any ALTER FUNCTION
I was not looking at ALTER FUNCTION but ALTER AGGREGATE. And I looked wrongly.
Sorry for that.
Generally, uppon rereading, I have to say that I am not very happy with the
decision that ANY triggers are fired from other places than the specific
triggers. That seams to be a rather dangerous/confusing route to me.
Especially because sometimes errors (permissions, duplicated names, etc) are
raised differently in ANY than in specific triggers now:
postgres=# ALTER AGGREGATE bar.array_agg_union3(anyarray) SET SCHEMA public;
NOTICE: when BEFORE, tag ALTER AGGREGATE, objectid <NULL>, schemaname <NULL>,
objectname <NULL>
ERROR: aggregate bar.array_agg_union3(anyarray) does not exist
postgres=# ALTER AGGREGATE public.array_agg_union3(anyarray) SET SCHEMA
public;
NOTICE: when BEFORE, tag ALTER AGGREGATE, objectid <NULL>, schemaname <NULL>,
objectname <NULL>
ERROR: function array_agg_union3(anyarray) is already in schema "public"
postgres=# ALTER AGGREGATE public.array_agg_union3(anyarray) SET SCHEMA bar;
NOTICE: when BEFORE, tag ALTER AGGREGATE, objectid <NULL>, schemaname <NULL>,
objectname <NULL>
NOTICE: when BEFORE, tag ALTER AGGREGATE, objectid 16415, schemaname public,
objectname array_agg_union3
NOTICE: when AFTER, tag ALTER AGGREGATE, objectid 16415, schemaname bar,
objectname array_agg_union3
NOTICE: when AFTER, tag ALTER AGGREGATE, objectid <NULL>, schemaname <NULL>,
objectname <NULL>
Andres
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
@ 2012-03-14 03:41 ` Robert Haas <[email protected]>
2012-03-14 08:27 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
0 siblings, 1 reply; 9629+ messages in thread
From: Robert Haas @ 2012-03-14 03:41 UTC (permalink / raw)
To: Andres Freund <[email protected]>; +Cc: Dimitri Fontaine <[email protected]>; pgsql-hackers
On Tue, Mar 13, 2012 at 5:06 PM, Andres Freund <[email protected]> wrote:
> Generally, uppon rereading, I have to say that I am not very happy with the
> decision that ANY triggers are fired from other places than the specific
> triggers. That seams to be a rather dangerous/confusing route to me.
I agree. I think that's a complete non-starter.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-14 03:41 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
@ 2012-03-14 08:27 ` Dimitri Fontaine <[email protected]>
2012-03-14 12:56 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
0 siblings, 1 reply; 9629+ messages in thread
From: Dimitri Fontaine @ 2012-03-14 08:27 UTC (permalink / raw)
To: Robert Haas <[email protected]>; +Cc: Andres Freund <[email protected]>; Dimitri Fontaine <[email protected]>; pgsql-hackers
Robert Haas <[email protected]> writes:
> On Tue, Mar 13, 2012 at 5:06 PM, Andres Freund <[email protected]> wrote:
>> Generally, uppon rereading, I have to say that I am not very happy with the
>> decision that ANY triggers are fired from other places than the specific
>> triggers. That seams to be a rather dangerous/confusing route to me.
>
> I agree. I think that's a complete non-starter.
Ok, well, let me react in 2 ways here:
A. it's very easy to change and will simplify the code
B. it's been done this way for good reasons (at the time)
Specifically, I've been asked to implement the feature of blocking all
and any DDL activity on a machine in a single command, and we don't have
support for triggers on all commands (remember shared objects).
Now, as I've completed support for all interesting commands the
discrepancy between what's supported in the ANY case and in the specific
command case has reduced. If you're saying to nothing, that's good news.
Also, when calling the user's procedure from the same place in case of an
ANY command trigger or a specific one it's then possible to just hand
them over the exact same set of info (object id, name, schema name).
Regards,
--
Dimitri Fontaine
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-14 03:41 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
2012-03-14 08:27 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
@ 2012-03-14 12:56 ` Robert Haas <[email protected]>
2012-03-14 21:33 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
0 siblings, 1 reply; 9629+ messages in thread
From: Robert Haas @ 2012-03-14 12:56 UTC (permalink / raw)
To: Dimitri Fontaine <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers
On Wed, Mar 14, 2012 at 4:27 AM, Dimitri Fontaine
<[email protected]> wrote:
> Also, when calling the user's procedure from the same place in case of an
> ANY command trigger or a specific one it's then possible to just hand
> them over the exact same set of info (object id, name, schema name).
Yes, I think that's an essential property of the system, here.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-14 03:41 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
2012-03-14 08:27 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-14 12:56 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
@ 2012-03-14 21:33 ` Dimitri Fontaine <[email protected]>
2012-03-15 12:03 ` Re: Command Triggers, patch v11 Thom Brown <[email protected]>
0 siblings, 1 reply; 9629+ messages in thread
From: Dimitri Fontaine @ 2012-03-14 21:33 UTC (permalink / raw)
To: Robert Haas <[email protected]>; +Cc: Dimitri Fontaine <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers
Robert Haas <[email protected]> writes:
>> Also, when calling the user's procedure from the same place in case of an
>> ANY command trigger or a specific one it's then possible to just hand
>> them over the exact same set of info (object id, name, schema name).
>
> Yes, I think that's an essential property of the system, here.
Ok, I've implemented that. No patch attached because I need to merge
with master again and I'm out to sleep now, it sometimes ring when being
on-call…
Curious people might have a look at my github repository where the
command_triggers branch is updated:
https://github.com/dimitri/postgres/compare/daf69e1e...e3714cb9e6
Regards,
--
Dimitri Fontaine
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-14 03:41 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
2012-03-14 08:27 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-14 12:56 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
2012-03-14 21:33 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
@ 2012-03-15 12:03 ` Thom Brown <[email protected]>
2012-03-15 14:52 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
0 siblings, 1 reply; 9629+ messages in thread
From: Thom Brown @ 2012-03-15 12:03 UTC (permalink / raw)
To: Dimitri Fontaine <[email protected]>; +Cc: Robert Haas <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers
On 14 March 2012 21:33, Dimitri Fontaine <[email protected]> wrote:
> Ok, I've implemented that. No patch attached because I need to merge
> with master again and I'm out to sleep now, it sometimes ring when being
> on-call…
>
> Curious people might have a look at my github repository where the
> command_triggers branch is updated:
Will you also be committing the trigger function variable changes
shortly? I don't wish to test anything prior to this as that will
involve a complete re-test from my side anyway.
--
Thom
^ permalink raw reply [nested|flat] 9629+ messages in thread
* Re: Command Triggers, patch v11
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 20:07 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-14 03:41 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
2012-03-14 08:27 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-14 12:56 ` Re: Command Triggers, patch v11 Robert Haas <[email protected]>
2012-03-14 21:33 ` Re: Command Triggers, patch v11 Dimitri Fontaine <[email protected]>
2012-03-15 12:03 ` Re: Command Triggers, patch v11 Thom Brown <[email protected]>
@ 2012-03-15 14:52 ` Dimitri Fontaine <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Dimitri Fontaine @ 2012-03-15 14:52 UTC (permalink / raw)
To: Thom Brown <[email protected]>; +Cc: Dimitri Fontaine <[email protected]>; Robert Haas <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers
Thom Brown <[email protected]> writes:
> Will you also be committing the trigger function variable changes
> shortly? I don't wish to test anything prior to this as that will
> involve a complete re-test from my side anyway.
It's on its way, I had to spend time elsewhere, sorry about that. With
some luck I can post a intermediate patch later this evening limited to
PLpgSQL support for your testing.
Regards,
--
Dimitri Fontaine
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-05 18:58 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-05 18:58 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH 5/5] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
* [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap.
@ 2026-03-09 10:35 Antonin Houska <[email protected]>
0 siblings, 0 replies; 9629+ messages in thread
From: Antonin Houska @ 2026-03-09 10:35 UTC (permalink / raw)
It should make the copying more efficient. Besides that, buffer access
strategy is involved this way.
This diff reverts the changes done in previous parts of the series to
reform_and_rewrite_tuple() and introduces a separate function
heap_insert_for_repack().
---
src/backend/access/heap/heapam_handler.c | 110 +++++++++++++++--------
1 file changed, 72 insertions(+), 38 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index f4c62e5bc7a..50ddee9d8c8 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -49,6 +49,11 @@
static void reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate);
+static void heap_insert_for_repack(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull,
+ BulkInsertState bistate);
+static HeapTuple reform_tuple(HeapTuple tuple, Relation OldHeap,
+ Relation NewHeap, Datum *values, bool *isnull);
static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
HeapTuple tuple,
@@ -694,6 +699,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
double *tups_recently_dead)
{
RewriteState rwstate = NULL;
+ BulkInsertState bistate = NULL;
IndexScanDesc indexScan;
TableScanDesc tableScan;
HeapScanDesc heapScan;
@@ -729,6 +735,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
if (!concurrent)
rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin,
*xid_cutoff, *multi_cutoff);
+ else
+ bistate = GetBulkInsertState();
/* Set up sorting if wanted */
@@ -958,8 +966,12 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
};
int64 ct_val[2];
- reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
- values, isnull, rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
+ values, isnull, rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
/*
* In indexscan mode and also VACUUM FULL, report increase in
@@ -1007,10 +1019,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
break;
n_tuples += 1;
- reform_and_rewrite_tuple(tuple,
- OldHeap, NewHeap,
- values, isnull,
- rwstate);
+ if (!concurrent)
+ reform_and_rewrite_tuple(tuple,
+ OldHeap, NewHeap,
+ values, isnull,
+ rwstate);
+ else
+ heap_insert_for_repack(tuple, OldHeap, NewHeap,
+ values, isnull, bistate);
+
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
@@ -1022,6 +1039,8 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Write out any remaining tuples, and fsync if needed */
if (rwstate)
end_heap_rewrite(rwstate);
+ if (bistate)
+ FreeBulkInsertState(bistate);
/* Clean up */
pfree(values);
@@ -2414,56 +2433,71 @@ heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
* SET WITHOUT OIDS.
*
* So, we must reconstruct the tuple from component Datums.
- *
- * If rwstate=NULL, use simple_heap_insert() instead of rewriting - in that
- * case we still need to deform/form the tuple. TODO Shouldn't we rename the
- * function, as might not do any rewrite?
*/
static void
reform_and_rewrite_tuple(HeapTuple tuple,
Relation OldHeap, Relation NewHeap,
Datum *values, bool *isnull, RewriteState rwstate)
+{
+ HeapTuple copiedTuple;
+
+ copiedTuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ /* The heap rewrite module does the rest */
+ rewrite_heap_tuple(rwstate, tuple, copiedTuple);
+
+ heap_freetuple(copiedTuple);
+}
+
+/*
+ * Insert tuple when processing REPACK CONCURRENTLY.
+ *
+ * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
+ * difficult to do the same in the catch-up phase (as the logical
+ * decoding does not provide us with sufficient visibility
+ * information). Thus we must use heap_insert() both during the
+ * catch-up and here.
+ *
+ * We pass the NO_LOGICAL flag to heap_insert() in order to skip logical
+ * decoding: as soon as REPACK CONCURRENTLY swaps the relation files, it drops
+ * this relation, so no logical replication subscription should need the data.
+ *
+ * BulkInsertState is used because many tuples are inserted in the typical
+ * case.
+ */
+static void
+heap_insert_for_repack(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull, BulkInsertState bistate)
+{
+ tuple = reform_tuple(tuple, OldHeap, NewHeap, values, isnull);
+
+ heap_insert(NewHeap, tuple, GetCurrentCommandId(true),
+ HEAP_INSERT_NO_LOGICAL, bistate);
+
+ heap_freetuple(tuple);
+}
+
+/*
+ * Deform tuple, set values of dropped columns to NULL, form a new tuple and
+ * return it.
+ */
+static HeapTuple
+reform_tuple(HeapTuple tuple, Relation OldHeap, Relation NewHeap,
+ Datum *values, bool *isnull)
{
TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
TupleDesc newTupDesc = RelationGetDescr(NewHeap);
- HeapTuple copiedTuple;
int i;
heap_deform_tuple(tuple, oldTupDesc, values, isnull);
- /* Be sure to null out any dropped columns */
for (i = 0; i < newTupDesc->natts; i++)
{
if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
isnull[i] = true;
}
- copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
-
- if (rwstate)
- /* The heap rewrite module does the rest */
- rewrite_heap_tuple(rwstate, tuple, copiedTuple);
- else
- {
- /*
- * Insert tuple when processing REPACK CONCURRENTLY.
- *
- * rewriteheap.c is not used in the CONCURRENTLY case because it'd be
- * difficult to do the same in the catch-up phase (as the logical
- * decoding does not provide us with sufficient visibility
- * information). Thus we must use heap_insert() both during the
- * catch-up and here.
- *
- * The following is like simple_heap_insert() except that we pass the
- * flag to skip logical decoding: as soon as REPACK CONCURRENTLY swaps
- * the relation files, it drops this relation, so no logical
- * replication subscription should need the data.
- */
- heap_insert(NewHeap, copiedTuple, GetCurrentCommandId(true),
- HEAP_INSERT_NO_LOGICAL, NULL);
- }
-
- heap_freetuple(copiedTuple);
+ return heap_form_tuple(newTupDesc, values, isnull);
}
/*
--
2.47.3
--k4rn5ob3nx2ufkij--
^ permalink raw reply [nested|flat] 9629+ messages in thread
end of thread, other threads:[~2026-03-09 10:35 UTC | newest]
Thread overview: 9629+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2012-03-13 11:22 Re: Command Triggers, patch v11 Andres Freund <[email protected]>
2012-03-13 15:14 ` Alvaro Herrera <[email protected]>
2012-03-13 20:07 ` Dimitri Fontaine <[email protected]>
2012-03-13 21:06 ` Andres Freund <[email protected]>
2012-03-14 03:41 ` Robert Haas <[email protected]>
2012-03-14 08:27 ` Dimitri Fontaine <[email protected]>
2012-03-14 12:56 ` Robert Haas <[email protected]>
2012-03-14 21:33 ` Dimitri Fontaine <[email protected]>
2012-03-15 12:03 ` Thom Brown <[email protected]>
2012-03-15 14:52 ` Dimitri Fontaine <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-05 18:58 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH 5/5] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
2026-03-09 10:35 [PATCH v40 4/4] Use BulkInsertState when copying data to the new heap. Antonin Houska <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox