pg.ddx.io pgsql-sql@postgresql.org mailing list archive
help / color / mirror / Atom feedProblem with large tuples.
5+ messages / 4 participants
[nested] [flat]
* Problem with large tuples.
@ 2000-01-21 15:05 Patrik Kudo <kudo@partitur.se>
0 siblings, 1 reply; 5+ messages in thread
From: Patrik Kudo @ 2000-01-21 15:05 UTC (permalink / raw)
To: pgsql-sql
I've noticed that some of my tables with large text tuples have
problem when these exceed some certain size. I know about the
8k tuplesize limit, but it seems the problem appears earlier
than that. This is what I've been able to recreate:
CREATE TABLE sizetest (id int, txt text);
INSERT INTO sizetest (id, txt) VALUES (1, '...'); // ... = string of
4000 chars
vacuum analyze; // <-- works nicely
INSERT INTO sizetest (id, txt) VALUES (2, '...'); // ... = string of
4100 chars
vacuum analyze;
ERROR: Tuple is too big: size 8152, max size 8140
How come the insert works while the vacuum fails?
Setup: Postgres 6.5.2, FreeBSD 3.3
/Patrik Kudo
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: [SQL] Problem with large tuples.
@ 2000-01-21 15:38 Dirk Lutzebaeck <lutzeb@aeccom.com>
parent: Patrik Kudo <kudo@partitur.se>
0 siblings, 1 reply; 5+ messages in thread
From: Dirk Lutzebaeck @ 2000-01-21 15:38 UTC (permalink / raw)
To: Patrik Kudo <kudo@partitur.se>; +Cc: pgsql-sql
Patrik Kudo writes:
> I've noticed that some of my tables with large text tuples have
> problem when these exceed some certain size. I know about the
> 8k tuplesize limit, but it seems the problem appears earlier
> than that. This is what I've been able to recreate:
>
> CREATE TABLE sizetest (id int, txt text);
> INSERT INTO sizetest (id, txt) VALUES (1, '...'); // ... = string of
> 4000 chars
> vacuum analyze; // <-- works nicely
> INSERT INTO sizetest (id, txt) VALUES (2, '...'); // ... = string of
> 4100 chars
> vacuum analyze;
> ERROR: Tuple is too big: size 8152, max size 8140
>
> How come the insert works while the vacuum fails?
It's a bug. You need this patch against 6.5.3:
inline patch 1: src/backend/commands/vacuum.c (+13/-2 lines) download inline-1.patch
*** src/backend/commands/vacuum.c.orig Wed Aug 25 08:01:45 1999
--- src/backend/commands/vacuum.c Tue Jan 4 12:15:17 2000
***************
*** 2405,2414 ****
stup = heap_formtuple(sd->rd_att, values, nulls);
/* ----------------
! * insert the tuple in the relation and get the tuple's oid.
* ----------------
*/
! heap_insert(sd, stup);
pfree(DatumGetPointer(values[3]));
pfree(DatumGetPointer(values[4]));
pfree(stup);
--- 2405,2425 ----
stup = heap_formtuple(sd->rd_att, values, nulls);
/* ----------------
! * Watch out for oversize tuple, which can happen if
! * both of the saved data values are long.
! * Our fallback strategy is just to not store the
! * pg_statistic tuple at all in that case. (We could
! * replace the values by NULLs and still store the
! * numeric stats, but presently selfuncs.c couldn't
! * do anything useful with that case anyway.)
* ----------------
*/
! if (MAXALIGN(stup->t_len) <= MaxTupleSize)
! {
! /* OK to store tuple */
! heap_insert(sd, stup);
! }
!
pfree(DatumGetPointer(values[3]));
pfree(DatumGetPointer(values[4]));
pfree(stup);
After patching you should remove your statistics with
'DELETE FROM pg_statistic';
You might also need to patch vio.c. Get it from the latest snapshot...
Go back in the pgsql-bugs archive to read more about this problem. It
was discussed some weeks ago...
Dirk
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: [SQL] Problem with large tuples.
@ 2000-01-25 11:29 Palle Girgensohn <girgen@partitur.se>
parent: Dirk Lutzebaeck <lutzeb@aeccom.com>
0 siblings, 1 reply; 5+ messages in thread
From: Palle Girgensohn @ 2000-01-25 11:29 UTC (permalink / raw)
To: Dirk Lutzebaeck <lutzeb@aeccom.com>; +Cc: Patrik Kudo <kudo@partitur.se>; pgsql-sql
Dirk Lutzebaeck wrote:
>
...
> You might also need to patch vio.c. Get it from the latest snapshot...
vio.c? Can't find it the 6.5.3 sources. I guess that means I don't need to patch it, right? ;)
/Palle
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: [SQL] Problem with large tuples.
@ 2000-01-25 17:18 Tom Lane <tgl@sss.pgh.pa.us>
parent: Palle Girgensohn <girgen@partitur.se>
0 siblings, 1 reply; 5+ messages in thread
From: Tom Lane @ 2000-01-25 17:18 UTC (permalink / raw)
To: Palle Girgensohn <girgen@partitur.se>; +Cc: Dirk Lutzebaeck <lutzeb@aeccom.com>; Patrik Kudo <kudo@partitur.se>; pgsql-sql
Palle Girgensohn <girgen@partitur.se> writes:
> Dirk Lutzebaeck wrote:
>> You might also need to patch vio.c. Get it from the latest snapshot...
> vio.c? Can't find it the 6.5.3 sources. I guess that means I don't need to patch it, right? ;)
I think he meant src/backend/access/heap/hio.c; there's a patch in there
that defends against adding an oversize tuple to a relation, IIRC.
regards, tom lane
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: [SQL] Problem with large tuples.
@ 2000-01-26 00:26 Palle Girgensohn <girgen@partitur.se>
parent: Tom Lane <tgl@sss.pgh.pa.us>
0 siblings, 0 replies; 5+ messages in thread
From: Palle Girgensohn @ 2000-01-26 00:26 UTC (permalink / raw)
To: Tom Lane <tgl@sss.pgh.pa.us>; +Cc: Dirk Lutzebaeck <lutzeb@aeccom.com>; Patrik Kudo <kudo@partitur.se>; pgsql-sql
Tom Lane wrote:
>
> I think he meant src/backend/access/heap/hio.c; there's a patch in there
> that defends against adding an oversize tuple to a relation, IIRC.
Thanks. I applied the patch.
/Palle
^ permalink raw reply [nested|flat] 5+ messages in thread
end of thread, other threads:[~2000-01-26 00:26 UTC | newest]
Thread overview: 5+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2000-01-21 15:05 Problem with large tuples. Patrik Kudo <kudo@partitur.se>
2000-01-21 15:38 ` Dirk Lutzebaeck <lutzeb@aeccom.com>
2000-01-25 11:29 ` Palle Girgensohn <girgen@partitur.se>
2000-01-25 17:18 ` Tom Lane <tgl@sss.pgh.pa.us>
2000-01-26 00:26 ` Palle Girgensohn <girgen@partitur.se>
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox