From: Andrey Lepikhov <a.lepikhov@postgrespro.ru>
To: Masahiko Sawada <sawada.mshk@gmail.com>
Cc: PostgreSQL-Dev <pgsql-hackers@postgresql.org>
Subject: Re: [WIP] [B-Tree] Retail IndexTuple deletion
Date: Wed, 15 Aug 2018 14:28:43 +0500
Message-ID: <247f4455-1ca4-b246-4ebf-283fca14ce03@postgrespro.ru> (raw)
In-Reply-To: <CAD21AoAn5n0HupuQACcmhvkvn9DDY92VM7pALqFxioutR2VWJQ@mail.gmail.com>
References: <425db134-8bba-005c-b59d-56e50de3b41e@postgrespro.ru>
<CAH2-WzmpjdeU77aJf1gtte3QhWrnwMffx3pPJQpEDHA1zbx3gw@mail.gmail.com>
<e4cec53a-ab12-4858-d9d8-5fac9948a4c7@postgrespro.ru>
<CAH2-Wz=mkrkQJJxHjDgiTv25yk0yqrx0EOn5exAHvPyQRR6iNQ@mail.gmail.com>
<9a008ab3-f591-e5e9-8e03-c0dd1dc13579@postgrespro.ru>
<CAL-rCA0BkRJgEfkQdDkii6kt0L-_8xcYqraaSmrrn_jzTFx-Pg@mail.gmail.com>
<2ada5790-dffa-a41a-ef84-10f60a525ba6@postgrespro.ru>
<CAH2-Wz=mnW+8+oZ9vh45+WEzt4dinYrz5f5acqn8mRjVY=YXqg@mail.gmail.com>
<CAH2-WzmPf0pmnS4go9+MmqQMA0MijsZ9XR+zdMy9J2gkEzadxA@mail.gmail.com>
<7598602c-f329-a6aa-00e6-e3335ed06048@postgrespro.ru>
<CAH2-Wzm2SG40Hp2nYMzqkKXz3-B9bxSM1N02bGNkft4djwYm7A@mail.gmail.com>
<CAD21AoApVGFf3q7WV5FuFzHDRtew9+fHHdCyOkk1uG+XG_6OKw@mail.gmail.com>
<CAH2-Wz=MkEWanC4WPTAwOdiHB7VpetgsY8i1qbYG9m-XY-PuTg@mail.gmail.com>
<f49bb262-d246-829d-f835-3950ddac503c@postgrespro.ru>
<CAD21AoAn5n0HupuQACcmhvkvn9DDY92VM7pALqFxioutR2VWJQ@mail.gmail.com>
15.08.2018 12:17, Masahiko Sawada пишет:
> On Tue, Aug 7, 2018 at 4:19 PM, Andrey Lepikhov
> <a.lepikhov@postgrespro.ru> wrote:
>> Hi,
>> I wrote a background worker (hcleaner) to demonstrate application of Retail
>> IndexTuple deletion (see patch at attachment).
>
> The patch doesn't seem to have the hcleaner code. Could you share it?
I appreciate you pointing out my mistake.
Attachment contains full patch set: indexTuple retail deletion, ordering
b-tree tuples by tid (provided by Peter Geoghean) and background cleaner.
In this version of background worker you can show state of the hcleaner
at the 'pg_stat_progress_cleaner' view, like VACUUM.
unlike the previous version, hcleaner check presence a block in memory
before cleanup (see RBM_NORMAL_NO_READ mode at ReadBufferExtended()
call) and do not read blocks from a disk storage (only on shutdown after
SIGTERM signal catch).
For feature demonstration you can use simple test (autovacuum = off):
pgbench -i -s 1 && psql -c $"CREATE INDEX pgbench_accounts_ext ON
public.pgbench_accounts USING btree (abalance);" && pgbench -t <n> -c 20
-j 8 -f test.pgb
where test.pgb is:
-------
\set aid random(1, 100000 * :scale)
\set delta random(-5000, 5000)
UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
-------
My workstation shows:
| master | patched |
n |HEAP size | INDEX size | HEAP size | INDEX size |
---------------|-------------------------------------|
2e3 | 13 MB | 2.3 MB | 13 MB | 2.3 MB |
2e4 | 14 MB | 2.7 MB | 13 MB | 2.7 MB |
2e5 | 14 MB | 8.0 MB | 14 MB | 4.8 MB |
2e6 | 61 MB | 58. MB | 14 MB | 6.7 MB |
where HEAP size - size of 'pgbench_accounts' relation; INDEX size - size
of 'pgbench_accounts_ext' index relation.
It is demonstrates a relation 'blowing' problem and influence of
hcleaner in an excessive manner.
Some problem is regression tests modification, because hcleaner makes
physical order of tuples in relations unpredictable.
>
> Regards,
>
> --
> Masahiko Sawada
> NIPPON TELEGRAPH AND TELEPHONE CORPORATION
> NTT Open Source Software Center
>
--
Andrey Lepikhov
Postgres Professional
https://postgrespro.com
The Russian Postgres Company
Attachments:download all (5 files, .tar.gz)0001-Retail-IndexTuple-Deletion-Access-Method.patch (text/x-patch, 12.0K) download
From 279739393a264b2d016afee3278e4afba655ce60 Mon Sep 17 00:00:00 2001
From: "Andrey V. Lepikhov" <a.lepikhov@postgrespro.ru>
Date: Wed, 15 Aug 2018 13:34:17 +0500
Subject: [PATCH 1/5] Retail-IndexTuple-Deletion-Access-Method
---
contrib/bloom/blutils.c | 1 +
src/backend/access/brin/brin.c | 1 +
src/backend/access/gin/ginutil.c | 1 +
src/backend/access/gist/gist.c | 1 +
src/backend/access/hash/hash.c | 1 +
src/backend/access/index/indexam.c | 15 +++
src/backend/access/nbtree/nbtree.c | 138 +++++++++++++++++++++++++++
src/backend/access/spgist/spgutils.c | 1 +
src/include/access/amapi.h | 6 ++
src/include/access/genam.h | 18 ++++
src/include/access/nbtree.h | 4 +
11 files changed, 187 insertions(+)
diff --git a/contrib/bloom/blutils.c b/contrib/bloom/blutils.cindex 6b2b9e3742..96f1d47c70 100644--- a/contrib/bloom/blutils.c+++ b/contrib/bloom/blutils.c@@ -126,6 +126,7 @@ blhandler(PG_FUNCTION_ARGS)
amroutine->ambuild = blbuild;
amroutine->ambuildempty = blbuildempty;
amroutine->aminsert = blinsert;
+ amroutine->amtargetdelete = NULL;
amroutine->ambulkdelete = blbulkdelete;
amroutine->amvacuumcleanup = blvacuumcleanup;
amroutine->amcanreturn = NULL;
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.cindex e95fbbcea7..a0e06bd868 100644--- a/src/backend/access/brin/brin.c+++ b/src/backend/access/brin/brin.c@@ -103,6 +103,7 @@ brinhandler(PG_FUNCTION_ARGS)
amroutine->ambuild = brinbuild;
amroutine->ambuildempty = brinbuildempty;
amroutine->aminsert = brininsert;
+ amroutine->amtargetdelete = NULL;
amroutine->ambulkdelete = brinbulkdelete;
amroutine->amvacuumcleanup = brinvacuumcleanup;
amroutine->amcanreturn = NULL;
diff --git a/src/backend/access/gin/ginutil.c b/src/backend/access/gin/ginutil.cindex 0a32182dd7..acf14e7bec 100644--- a/src/backend/access/gin/ginutil.c+++ b/src/backend/access/gin/ginutil.c@@ -58,6 +58,7 @@ ginhandler(PG_FUNCTION_ARGS)
amroutine->ambuild = ginbuild;
amroutine->ambuildempty = ginbuildempty;
amroutine->aminsert = gininsert;
+ amroutine->amtargetdelete = NULL;
amroutine->ambulkdelete = ginbulkdelete;
amroutine->amvacuumcleanup = ginvacuumcleanup;
amroutine->amcanreturn = NULL;
diff --git a/src/backend/access/gist/gist.c b/src/backend/access/gist/gist.cindex 8a42effdf7..d7a53d2ca9 100644--- a/src/backend/access/gist/gist.c+++ b/src/backend/access/gist/gist.c@@ -80,6 +80,7 @@ gisthandler(PG_FUNCTION_ARGS)
amroutine->ambuild = gistbuild;
amroutine->ambuildempty = gistbuildempty;
amroutine->aminsert = gistinsert;
+ amroutine->amtargetdelete = NULL;
amroutine->ambulkdelete = gistbulkdelete;
amroutine->amvacuumcleanup = gistvacuumcleanup;
amroutine->amcanreturn = gistcanreturn;
diff --git a/src/backend/access/hash/hash.c b/src/backend/access/hash/hash.cindex 0002df30c0..5fb32d6eba 100644--- a/src/backend/access/hash/hash.c+++ b/src/backend/access/hash/hash.c@@ -76,6 +76,7 @@ hashhandler(PG_FUNCTION_ARGS)
amroutine->ambuild = hashbuild;
amroutine->ambuildempty = hashbuildempty;
amroutine->aminsert = hashinsert;
+ amroutine->amtargetdelete = NULL;
amroutine->ambulkdelete = hashbulkdelete;
amroutine->amvacuumcleanup = hashvacuumcleanup;
amroutine->amcanreturn = NULL;
diff --git a/src/backend/access/index/indexam.c b/src/backend/access/index/indexam.cindex 22b5cc921f..9ebeb789fb 100644--- a/src/backend/access/index/indexam.c+++ b/src/backend/access/index/indexam.c@@ -730,6 +730,21 @@ index_getbitmap(IndexScanDesc scan, TIDBitmap *bitmap)
return ntids;
}
+IndexTargetDeleteResult *+index_target_delete(IndexTargetDeleteInfo *info,+ IndexTargetDeleteResult *stats,+ Datum *values,+ bool *isnull)+{+ Relation indexRelation = info->indexRelation;++ RELATION_CHECKS;++ CHECK_REL_PROCEDURE(amtargetdelete);++ return indexRelation->rd_amroutine->amtargetdelete(info, stats, values, isnull);+}+
/* ----------------
* index_bulk_delete - do mass deletion of index entries
*
diff --git a/src/backend/access/nbtree/nbtree.c b/src/backend/access/nbtree/nbtree.cindex e8725fbbe1..44f4d0dc9f 100644--- a/src/backend/access/nbtree/nbtree.c+++ b/src/backend/access/nbtree/nbtree.c@@ -34,7 +34,9 @@
#include "storage/smgr.h"
#include "utils/builtins.h"
#include "utils/index_selfuncs.h"
+#include "utils/lsyscache.h"
#include "utils/memutils.h"
+#include "utils/syscache.h"
/* Working state needed by btvacuumpage */
@@ -127,6 +129,7 @@ bthandler(PG_FUNCTION_ARGS)
amroutine->ambuild = btbuild;
amroutine->ambuildempty = btbuildempty;
amroutine->aminsert = btinsert;
+ amroutine->amtargetdelete = bttargetdelete;
amroutine->ambulkdelete = btbulkdelete;
amroutine->amvacuumcleanup = btvacuumcleanup;
amroutine->amcanreturn = btcanreturn;
@@ -886,6 +889,141 @@ btbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
return stats;
}
+static int+tid_list_search(ItemPointer tid, ItemPointer tid_list, int ntid)+{+ for (int i = 0; i < ntid; i++)+ if (ItemPointerEquals(tid, &(tid_list[i])))+ return i;+ return -1;+}++IndexTargetDeleteResult*+bttargetdelete(IndexTargetDeleteInfo *info,+ IndexTargetDeleteResult *stats,+ Datum *values,+ bool *isnull)+{+ Relation irel = info->indexRelation;+ Relation hrel = info->heapRelation;+ ScanKey skey;+ int keysCount = IndexRelationGetNumberOfKeyAttributes(irel);+ BTStack stack;+ Buffer buf;+ Page page;+ BTPageOpaque opaque;+ OffsetNumber offnum;+ int ndeletable = 0;+ OffsetNumber deletable[MaxOffsetNumber];+ IndexTuple itup;++ if (stats == NULL)+ stats = (IndexTargetDeleteResult *) palloc0(sizeof(IndexTargetDeleteResult));++ itup = index_form_tuple(RelationGetDescr(irel), values, isnull);+ skey = _bt_mkscankey(irel, itup);++ /* Descend the tree and position ourselves on the target leaf page. */+ stack = _bt_search(irel, keysCount, skey, false, &buf, BT_READ, NULL);+ _bt_freestack(stack);++ /* To prepare tuple entries search across index pages */+ Assert(BufferIsValid(buf));+ offnum = _bt_binsrch(irel, buf, keysCount, skey, false);+ page = BufferGetPage(buf);+ _bt_checkpage(irel, buf);+ opaque = (BTPageOpaque) PageGetSpecialPointer(page);++ for (;;)+ {+ int32 cmpval;+ ItemId itemid;+ IndexTuple itup;+ int pos;++ /* Switch to the next page */+ if (P_IGNORE(opaque) || (offnum > PageGetMaxOffsetNumber(page)))+ {+ /*+ * Before unlocking index page we need to delete+ * all currently found tuples+ */+ if (ndeletable > 0)+ {+ LockBuffer(buf, BUFFER_LOCK_UNLOCK);+ LockBufferForCleanup(buf);++ _bt_delitems_delete(irel, buf, deletable, ndeletable, hrel);++ stats->tuples_removed += ndeletable;+ ndeletable = 0;+ }++ /*+ * Check for end-of-index+ */+ if (P_RIGHTMOST(opaque))+ /* it is rightmost leaf */+ break;++ /*+ * Switch to the next index page+ */+ buf = _bt_relandgetbuf(irel, buf, opaque->btpo_next, BT_READ);+ page = BufferGetPage(buf);+ _bt_checkpage(irel, buf);+ opaque = (BTPageOpaque) PageGetSpecialPointer(page);+ offnum = P_FIRSTDATAKEY(opaque);+ continue;+ }++ /*+ * This index entry satisfied to the scan key?+ */+ cmpval = _bt_compare(irel, keysCount, skey, page, offnum);++ if (cmpval != 0)+ /* End of index entries, satisfied to the scan key */+ break;++ /*+ * To load index tuple and look for matches in the TID list of+ * dead heap tuples+ */+ itemid = PageGetItemId(page, offnum);+ itup = (IndexTuple) PageGetItem(page, itemid);+ pos = tid_list_search(&(itup->t_tid), info->dead_tuples, info->num_dead_tuples);++ if ((pos >= 0) && (!info->found_dead_tuples[pos]))+ {+ /* index entry for TID of dead tuple is found */+ deletable[ndeletable++] = offnum;+ info->found_dead_tuples[pos] = true;+ }++ offnum = OffsetNumberNext(offnum);+ }++ /*+ * Delete all found index tuples+ */+ if (ndeletable > 0)+ {+ LockBuffer(buf, BUFFER_LOCK_UNLOCK);+ LockBufferForCleanup(buf);++ _bt_delitems_delete(irel, buf, deletable, ndeletable, hrel);++ stats->tuples_removed += ndeletable;+ }++ /* Release scan key, unpin and unlock buffer */+ _bt_freeskey(skey);+ _bt_relbuf(irel, buf);++ return stats;+}+
/*
* Post-VACUUM cleanup.
*
diff --git a/src/backend/access/spgist/spgutils.c b/src/backend/access/spgist/spgutils.cindex 4a9b5da268..325728122b 100644--- a/src/backend/access/spgist/spgutils.c+++ b/src/backend/access/spgist/spgutils.c@@ -56,6 +56,7 @@ spghandler(PG_FUNCTION_ARGS)
amroutine->ambuild = spgbuild;
amroutine->ambuildempty = spgbuildempty;
amroutine->aminsert = spginsert;
+ amroutine->amtargetdelete = NULL;
amroutine->ambulkdelete = spgbulkdelete;
amroutine->amvacuumcleanup = spgvacuumcleanup;
amroutine->amcanreturn = spgcanreturn;
diff --git a/src/include/access/amapi.h b/src/include/access/amapi.hindex 14526a6bb2..497b54e8a8 100644--- a/src/include/access/amapi.h+++ b/src/include/access/amapi.h@@ -76,6 +76,11 @@ typedef bool (*aminsert_function) (Relation indexRelation,
IndexUniqueCheck checkUnique,
struct IndexInfo *indexInfo);
+/* target delete */+typedef IndexTargetDeleteResult *(*amtargetdelete_function) (IndexTargetDeleteInfo *info,+ IndexTargetDeleteResult *stats,+ Datum *values,+ bool *isnull);
/* bulk delete */
typedef IndexBulkDeleteResult *(*ambulkdelete_function) (IndexVacuumInfo *info,
IndexBulkDeleteResult *stats,
@@ -207,6 +212,7 @@ typedef struct IndexAmRoutine
ambuild_function ambuild;
ambuildempty_function ambuildempty;
aminsert_function aminsert;
+ amtargetdelete_function amtargetdelete;
ambulkdelete_function ambulkdelete;
amvacuumcleanup_function amvacuumcleanup;
amcanreturn_function amcanreturn; /* can be NULL */
diff --git a/src/include/access/genam.h b/src/include/access/genam.hindex 24c720bf42..79b386eebd 100644--- a/src/include/access/genam.h+++ b/src/include/access/genam.h@@ -51,6 +51,15 @@ typedef struct IndexVacuumInfo
BufferAccessStrategy strategy; /* access strategy for reads */
} IndexVacuumInfo;
+typedef struct IndexTargetDeleteInfo+{+ Relation heapRelation;+ Relation indexRelation; /* the index being vacuumed */+ int num_dead_tuples;+ ItemPointer dead_tuples;+ bool* found_dead_tuples;+} IndexTargetDeleteInfo;+
/*
* Struct for statistics returned by ambulkdelete and amvacuumcleanup
*
@@ -79,6 +88,11 @@ typedef struct IndexBulkDeleteResult
BlockNumber pages_free; /* # pages available for reuse */
} IndexBulkDeleteResult;
+typedef struct IndexTargetDeleteResult+{+ int tuples_removed; /* # removed during vacuum operation */+} IndexTargetDeleteResult;+
/* Typedef for callback function to determine if a tuple is bulk-deletable */
typedef bool (*IndexBulkDeleteCallback) (ItemPointer itemptr, void *state);
@@ -163,6 +177,10 @@ extern HeapTuple index_fetch_heap(IndexScanDesc scan);
extern HeapTuple index_getnext(IndexScanDesc scan, ScanDirection direction);
extern int64 index_getbitmap(IndexScanDesc scan, TIDBitmap *bitmap);
+extern IndexTargetDeleteResult *index_target_delete(IndexTargetDeleteInfo *info,+ IndexTargetDeleteResult *stats,+ Datum *values,+ bool *isnull);
extern IndexBulkDeleteResult *index_bulk_delete(IndexVacuumInfo *info,
IndexBulkDeleteResult *stats,
IndexBulkDeleteCallback callback,
diff --git a/src/include/access/nbtree.h b/src/include/access/nbtree.hindex 04ecb4cbc0..3b9461c3b4 100644--- a/src/include/access/nbtree.h+++ b/src/include/access/nbtree.h@@ -508,6 +508,10 @@ extern void btparallelrescan(IndexScanDesc scan);
extern void btendscan(IndexScanDesc scan);
extern void btmarkpos(IndexScanDesc scan);
extern void btrestrpos(IndexScanDesc scan);
+extern IndexTargetDeleteResult *bttargetdelete(IndexTargetDeleteInfo *info,+ IndexTargetDeleteResult *stats,+ Datum *values,+ bool *isnull);
extern IndexBulkDeleteResult *btbulkdelete(IndexVacuumInfo *info,
IndexBulkDeleteResult *stats,
IndexBulkDeleteCallback callback,
--
2.17.1
0002-Quick-Vacuum-Strategy.patch (text/x-patch, 12.3K) download
From 5cf988490d653fea24aa06e1c36af90f912fee7d Mon Sep 17 00:00:00 2001
From: "Andrey V. Lepikhov" <a.lepikhov@postgrespro.ru>
Date: Wed, 15 Aug 2018 13:34:45 +0500
Subject: [PATCH 2/5] Quick-Vacuum-Strategy
---
src/backend/access/heap/heapam.c | 31 ++++++
src/backend/access/heap/pruneheap.c | 40 ++++++-
src/backend/commands/vacuumlazy.c | 161 +++++++++++++++++++++++++++-
src/backend/utils/misc/guc.c | 10 +-
src/include/access/heapam.h | 1 +
5 files changed, 234 insertions(+), 9 deletions(-)
diff --git a/src/backend/access/heap/heapam.c b/src/backend/access/heap/heapam.cindex 72395a50b8..dd033668b6 100644--- a/src/backend/access/heap/heapam.c+++ b/src/backend/access/heap/heapam.c@@ -9392,6 +9392,36 @@ heap_sync(Relation rel)
}
}
+/*+ * Mask DEAD tuples at a HEAP page+ *+ * We introduce this function for wal_consistency_checking satisfaction at+ * Hot Standby node.+ *+ * DEAD tuple on a master node has t_cid and t_infomask, which can not be+ * obtained by WAL records applying on a Hot Standby node.+ */+static void+mask_dead_tuples(Page page)+{+ OffsetNumber offnum;++ for (offnum = FirstOffsetNumber; offnum <= PageGetMaxOffsetNumber(page); offnum = OffsetNumberNext(offnum))+ {+ ItemId lp = PageGetItemId(page, offnum);+ char* htup_begin;++ if (!ItemIdIsDead(lp))+ continue;++ if (!ItemIdHasStorage(lp))+ continue;++ htup_begin = (char *) PageGetItem(page, lp);+ memset(htup_begin, MASK_MARKER, ItemIdGetLength(lp));+ }+}+
/*
* Mask a heap page before performing consistency checks on it.
*/
@@ -9405,6 +9435,7 @@ heap_mask(char *pagedata, BlockNumber blkno)
mask_page_hint_bits(page);
mask_unused_space(page);
+ mask_dead_tuples(page);
for (off = 1; off <= PageGetMaxOffsetNumber(page); off++)
{
diff --git a/src/backend/access/heap/pruneheap.c b/src/backend/access/heap/pruneheap.cindex c2f5343dac..ebbafe3275 100644--- a/src/backend/access/heap/pruneheap.c+++ b/src/backend/access/heap/pruneheap.c@@ -43,6 +43,9 @@ typedef struct
bool marked[MaxHeapTuplesPerPage + 1];
} PruneState;
+/* Parameter for target deletion strategy in lazy vacuum */+double target_index_deletion_factor = 0.01;+
/* Local functions */
static int heap_prune_chain(Relation relation, Buffer buffer,
OffsetNumber rootoffnum,
@@ -405,7 +408,7 @@ heap_prune_chain(Relation relation, Buffer buffer, OffsetNumber rootoffnum,
if (HeapTupleSatisfiesVacuum(&tup, OldestXmin, buffer)
== HEAPTUPLE_DEAD && !HeapTupleHeaderIsHotUpdated(htup))
{
- heap_prune_record_unused(prstate, rootoffnum);+ heap_prune_record_dead(prstate, rootoffnum);
HeapTupleHeaderAdvanceLatestRemovedXid(htup,
&prstate->latestRemovedXid);
ndeleted++;
@@ -580,8 +583,10 @@ heap_prune_chain(Relation relation, Buffer buffer, OffsetNumber rootoffnum,
*/
for (i = 1; (i < nchain) && (chainitems[i - 1] != latestdead); i++)
{
- heap_prune_record_unused(prstate, chainitems[i]);
ndeleted++;
+ if (chainitems[i] == latestdead)+ continue;+ heap_prune_record_unused(prstate, chainitems[i]);
}
/*
@@ -598,9 +603,28 @@ heap_prune_chain(Relation relation, Buffer buffer, OffsetNumber rootoffnum,
* redirect the root to the correct chain member.
*/
if (i >= nchain)
+ {+ if (rootoffnum != latestdead)+ {+ if (ItemIdIsNormal(rootlp))+ heap_prune_record_unused(prstate, latestdead);+ else+ {+ /*+ * We allow overlapping of redirected and dead items+ */+ heap_prune_record_redirect(prstate, rootoffnum, latestdead);+ heap_prune_record_dead(prstate, latestdead);+ }+ }
heap_prune_record_dead(prstate, rootoffnum);
+ }
else
+ {+ if (rootoffnum != latestdead)+ heap_prune_record_unused(prstate, latestdead);
heap_prune_record_redirect(prstate, rootoffnum, chainitems[i]);
+ }
}
else if (nchain < 2 && ItemIdIsRedirected(rootlp))
{
@@ -653,7 +677,12 @@ heap_prune_record_dead(PruneState *prstate, OffsetNumber offnum)
Assert(prstate->ndead < MaxHeapTuplesPerPage);
prstate->nowdead[prstate->ndead] = offnum;
prstate->ndead++;
- Assert(!prstate->marked[offnum]);++ /*+ * We suppress checking prstate->marked[offnum]. It is not the best idea,+ * but this is most simplistic way to enable Dead Redirecting by+ * overlapping Dead and Redirected states.+ */
prstate->marked[offnum] = true;
}
@@ -706,7 +735,10 @@ heap_page_prune_execute(Buffer buffer,
OffsetNumber off = *offnum++;
ItemId lp = PageGetItemId(page, off);
- ItemIdSetDead(lp);+ if (target_index_deletion_factor > 0)+ ItemIdMarkDead(lp);+ else+ ItemIdSetDead(lp);
}
/* Update all now-unused line pointers */
diff --git a/src/backend/commands/vacuumlazy.c b/src/backend/commands/vacuumlazy.cindex 5649a70800..9a625b8dba 100644--- a/src/backend/commands/vacuumlazy.c+++ b/src/backend/commands/vacuumlazy.c@@ -41,13 +41,16 @@
#include "access/heapam_xlog.h"
#include "access/htup_details.h"
#include "access/multixact.h"
+#include "access/nbtree.h"
#include "access/transam.h"
#include "access/visibilitymap.h"
#include "access/xlog.h"
+#include "catalog/index.h"
#include "catalog/storage.h"
#include "commands/dbcommands.h"
#include "commands/progress.h"
#include "commands/vacuum.h"
+#include "executor/executor.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "portability/instr_time.h"
@@ -139,7 +142,6 @@ typedef struct LVRelStats
bool lock_waiter_detected;
} LVRelStats;
-
/* A few variables that don't seem worth passing around as parameters */
static int elevel = -1;
@@ -156,6 +158,8 @@ static void lazy_scan_heap(Relation onerel, int options,
bool aggressive);
static void lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats);
static bool lazy_check_needs_freeze(Buffer buf, bool *hastup);
+static void quick_vacuum_index(Relation irel, Relation hrel,+ LVRelStats *vacrelstats);
static void lazy_vacuum_index(Relation indrel,
IndexBulkDeleteResult **stats,
LVRelStats *vacrelstats);
@@ -734,10 +738,17 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats,
/* Remove index entries */
for (i = 0; i < nindexes; i++)
- lazy_vacuum_index(Irel[i],+ {+ bool use_quick_strategy = (vacrelstats->num_dead_tuples/vacrelstats->old_live_tuples < target_index_deletion_factor);++ if (use_quick_strategy && (Irel[i]->rd_amroutine->amtargetdelete != NULL))+ quick_vacuum_index(Irel[i], onerel, vacrelstats);+ else+ lazy_vacuum_index(Irel[i],
&indstats[i],
vacrelstats);
+ }
/*
* Report that we are now vacuuming the heap. We also increase
* the number of index scans here; note that by using
@@ -1379,10 +1390,16 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats,
/* Remove index entries */
for (i = 0; i < nindexes; i++)
- lazy_vacuum_index(Irel[i],+ {+ bool use_quick_strategy = (vacrelstats->num_dead_tuples/vacrelstats->old_live_tuples < target_index_deletion_factor);++ if (use_quick_strategy && (Irel[i]->rd_amroutine->amtargetdelete != NULL))+ quick_vacuum_index(Irel[i], onerel, vacrelstats);+ else+ lazy_vacuum_index(Irel[i],
&indstats[i],
vacrelstats);
-+ }
/* Report that we are now vacuuming the heap */
hvp_val[0] = PROGRESS_VACUUM_PHASE_VACUUM_HEAP;
hvp_val[1] = vacrelstats->num_index_scans + 1;
@@ -1671,6 +1688,142 @@ lazy_check_needs_freeze(Buffer buf, bool *hastup)
return false;
}
+/*+ * Get tuple from heap for a scan key building.+ */+static HeapTuple+get_tuple_by_tid(Relation rel, ItemPointer tid)+{+ Buffer buffer;+ Page page;+ OffsetNumber offnum;+ ItemId lp;+ HeapTuple tuple;+ bool needLock = !RELATION_IS_LOCAL(rel);+ BlockNumber npages;++ if (needLock)+ LockRelationForExtension(rel, ExclusiveLock);+ npages = RelationGetNumberOfBlocks(rel);+ if (needLock)+ UnlockRelationForExtension(rel, ExclusiveLock);+ if (ItemPointerGetBlockNumber(tid) > npages)+ return NULL;++ buffer = ReadBufferExtended(rel, MAIN_FORKNUM, ItemPointerGetBlockNumber(tid), RBM_NORMAL, NULL);+ LockBuffer(buffer, BUFFER_LOCK_SHARE);++ page = (Page) BufferGetPage(buffer);+ offnum = ItemPointerGetOffsetNumber(tid);+ lp = PageGetItemId(page, offnum);++ /*+ * VACUUM Races: someone already remove the tuple from HEAP. Ignore it.+ */+ if (!ItemIdIsUsed(lp))+ {+ UnlockReleaseBuffer(buffer);+ return NULL;+ }+ /* Walk along the chain */+ while (!ItemIdHasStorage(lp))+ {+ offnum = ItemIdGetRedirect(lp);+ lp = PageGetItemId(page, offnum);+ Assert(ItemIdIsUsed(lp));+ }++ /* Form a tuple */+ tuple = palloc(sizeof(HeapTupleData));+ ItemPointerSet(&(tuple->t_self), BufferGetBlockNumber(buffer), offnum);+ tuple->t_tableOid = RelationGetRelid(rel);+ tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);+ tuple->t_len = ItemIdGetLength(lp);+ UnlockReleaseBuffer(buffer);+ return tuple;+}++/*+ * quick_vacuum_index() -- quick vacuum one index relation.+ *+ * Delete all the index entries pointing to tuples listed in+ * vacrelstats->dead_tuples.+ */+static void+quick_vacuum_index(Relation irel, Relation hrel,+ LVRelStats *vacrelstats)+{+ int tnum;+ bool* found = palloc0(vacrelstats->num_dead_tuples*sizeof(bool));+ IndexInfo* indexInfo = BuildIndexInfo(irel);+ EState* estate = CreateExecutorState();+ ExprContext* econtext = GetPerTupleExprContext(estate);+ ExprState* predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);+ TupleTableSlot* slot = MakeSingleTupleTableSlot(RelationGetDescr(hrel));++ IndexTargetDeleteResult stats;+ IndexTargetDeleteInfo ivinfo;++ ivinfo.indexRelation = irel;+ ivinfo.heapRelation = hrel;++ econtext->ecxt_scantuple = slot;++ /* Get tuple from heap */+ for (tnum = 0; tnum < vacrelstats->num_dead_tuples; tnum++)+ {+ HeapTuple tuple;+ Datum values[INDEX_MAX_KEYS];+ bool isnull[INDEX_MAX_KEYS];++ /* Index entry for the TID was deleted early */+ if (found[tnum])+ continue;++ /* Get a tuple from heap */+ if ((tuple = get_tuple_by_tid(hrel, &(vacrelstats->dead_tuples[tnum]))) == NULL)+ {+ /*+ * Tuple has 'not used' status.+ */+ found[tnum] = true;+ continue;+ }++ /*+ * Form values[] and isnull[] arrays from for index tuple+ * by heap tuple+ */+ MemoryContextReset(econtext->ecxt_per_tuple_memory);++ ExecStoreTuple(tuple, slot, InvalidBuffer, false);++ /*+ * In a partial index, ignore tuples that don't satisfy the+ * predicate.+ */+ if ((predicate != NULL) && (!ExecQual(predicate, econtext)))+ {+ found[tnum] = true;+ continue;+ }++ FormIndexDatum(indexInfo, slot, estate, values, isnull);++ /*+ * Make attempt to delete some index entries by one tree descent.+ * We use only a part of TID list, which contains not found TID's.+ */+ ivinfo.dead_tuples = &(vacrelstats->dead_tuples[tnum]);+ ivinfo.num_dead_tuples = vacrelstats->num_dead_tuples - tnum;+ ivinfo.found_dead_tuples = found + tnum;+ index_target_delete(&ivinfo, &stats, values, isnull);+ }++ pfree(found);+ ExecDropSingleTupleTableSlot(slot);+ FreeExecutorState(estate);+}
/*
* lazy_vacuum_index() -- vacuum one index relation.
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.cindex a88ea6cfc9..e9e953ece6 100644--- a/src/backend/utils/misc/guc.c+++ b/src/backend/utils/misc/guc.c@@ -3243,7 +3243,15 @@ static struct config_real ConfigureNamesReal[] =
0.1, 0.0, 100.0,
NULL, NULL, NULL
},
-+ {+ {"target_index_deletion_factor", PGC_SIGHUP, AUTOVACUUM,+ gettext_noop("Maximum number of vacuumed tuples as a fraction of reltuples where we can use target index vacuum strategy."),+ NULL+ },+ &target_index_deletion_factor,+ 0.01, 0.0, 1.0,+ NULL, NULL, NULL+ },
{
{"checkpoint_completion_target", PGC_SIGHUP, WAL_CHECKPOINTS,
gettext_noop("Time spent flushing dirty buffers during checkpoint, as fraction of checkpoint interval."),
diff --git a/src/include/access/heapam.h b/src/include/access/heapam.hindex ca5cad7497..c5a7bc8361 100644--- a/src/include/access/heapam.h+++ b/src/include/access/heapam.h@@ -100,6 +100,7 @@ extern Relation heap_openrv_extended(const RangeVar *relation,
typedef struct HeapScanDescData *HeapScanDesc;
typedef struct ParallelHeapScanDescData *ParallelHeapScanDesc;
+extern double target_index_deletion_factor;
/*
* HeapScanIsValid
* True iff the heap scan is valid.
--
2.17.1
0003-Make-all-nbtree-index-tuples-have-unique-keys.patch (text/x-patch, 104.1K) download
From e55cb5b5ed3507ae7e6be30f2637075eba76e42e Mon Sep 17 00:00:00 2001
From: "Andrey V. Lepikhov" <a.lepikhov@postgrespro.ru>
Date: Wed, 15 Aug 2018 13:35:14 +0500
Subject: [PATCH 3/5] Make-all-nbtree-index-tuples-have-unique-keys
---
contrib/amcheck/verify_nbtree.c | 256 +++++++++---
contrib/pageinspect/expected/btree.out | 2 +-
contrib/pgstattuple/expected/pgstattuple.out | 10 +-
src/backend/access/nbtree/README | 116 +++--
src/backend/access/nbtree/nbtinsert.c | 418 ++++++++++++-------
src/backend/access/nbtree/nbtpage.c | 8 +-
src/backend/access/nbtree/nbtsearch.c | 168 ++++++--
src/backend/access/nbtree/nbtsort.c | 63 ++-
src/backend/access/nbtree/nbtutils.c | 267 ++++++++++--
src/backend/access/nbtree/nbtxlog.c | 41 +-
src/backend/access/rmgrdesc/nbtdesc.c | 8 -
src/backend/storage/page/bufpage.c | 4 +-
src/backend/utils/sort/tuplesort.c | 13 +-
src/include/access/nbtree.h | 85 +++-
src/include/access/nbtxlog.h | 20 +-
src/test/regress/expected/join.out | 2 +-
16 files changed, 1064 insertions(+), 417 deletions(-)
diff --git a/contrib/amcheck/verify_nbtree.c b/contrib/amcheck/verify_nbtree.cindex a1438a2855..eb34b22c30 100644--- a/contrib/amcheck/verify_nbtree.c+++ b/contrib/amcheck/verify_nbtree.c@@ -45,6 +45,13 @@ PG_MODULE_MAGIC;
*/
#define InvalidBtreeLevel ((uint32) InvalidBlockNumber)
+/*+ * Convenience macro to get number of key attributes in tuple in low-context+ * fashion+ */+#define BTreeTupleGetNKeyAtts(itup, rel) \+ Min(IndexRelationGetNumberOfKeyAttributes(rel), BTreeTupleGetNAtts(itup, rel))+
/*
* State associated with verifying a B-Tree index
*
@@ -125,26 +132,30 @@ static void bt_check_every_level(Relation rel, Relation heaprel,
static BtreeLevel bt_check_level_from_leftmost(BtreeCheckState *state,
BtreeLevel level);
static void bt_target_page_check(BtreeCheckState *state);
-static ScanKey bt_right_page_check_scankey(BtreeCheckState *state);+static IndexTuple bt_right_page_check_tuple(BtreeCheckState *state);
static void bt_downlink_check(BtreeCheckState *state, BlockNumber childblock,
- ScanKey targetkey);+ ScanKey targetkey, ItemPointer scantid, int tupnkeyatts);
static void bt_downlink_missing_check(BtreeCheckState *state);
static void bt_tuple_present_callback(Relation index, HeapTuple htup,
Datum *values, bool *isnull,
bool tupleIsAlive, void *checkstate);
static inline bool offset_is_negative_infinity(BTPageOpaque opaque,
OffsetNumber offset);
+static inline bool invariant_l_offset(BtreeCheckState *state,+ int tupnkeyatts, ScanKey key, ItemPointer scantid,+ OffsetNumber upperbound);
static inline bool invariant_leq_offset(BtreeCheckState *state,
- ScanKey key,+ int tupnkeyatts, ScanKey key, ItemPointer scantid,
OffsetNumber upperbound);
-static inline bool invariant_geq_offset(BtreeCheckState *state,- ScanKey key,+static inline bool invariant_g_offset(BtreeCheckState *state,+ int tupnkeyatts, ScanKey key, ItemPointer scantid,
OffsetNumber lowerbound);
-static inline bool invariant_leq_nontarget_offset(BtreeCheckState *state,- Page other,- ScanKey key,- OffsetNumber upperbound);+static inline bool invariant_l_nontarget_offset(BtreeCheckState *state,+ Page other, int tupnkeyatts, ScanKey key,+ ItemPointer scantid, OffsetNumber upperbound);
static Page palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum);
+static inline ItemPointer BTreeTupleGetHeapTIDCareful(BtreeCheckState *state,+ IndexTuple itup, bool isleaf);
/*
* bt_index_check(index regclass, heapallindexed boolean)
@@ -834,8 +845,10 @@ bt_target_page_check(BtreeCheckState *state)
{
ItemId itemid;
IndexTuple itup;
- ScanKey skey;
size_t tupsize;
+ int tupnkeyatts;+ ScanKey skey;+ ItemPointer scantid;
CHECK_FOR_INTERRUPTS();
@@ -902,8 +915,17 @@ bt_target_page_check(BtreeCheckState *state)
if (offset_is_negative_infinity(topaque, offset))
continue;
- /* Build insertion scankey for current page offset */+ /*+ * Build insertion scankey for current page offset/tuple.+ *+ * As required by _bt_mkscankey(), track number of key attributes,+ * which is needed so that _bt_compare() calls handle truncated+ * attributes correctly. Never count non-key attributes in+ * non-truncated tuples as key attributes, though.+ */+ tupnkeyatts = BTreeTupleGetNKeyAtts(itup, state->rel);
skey = _bt_mkscankey(state->rel, itup);
+ scantid = BTreeTupleGetHeapTIDCareful(state, itup, P_ISLEAF(topaque));
/* Fingerprint leaf page tuples (those that point to the heap) */
if (state->heapallindexed && P_ISLEAF(topaque) && !ItemIdIsDead(itemid))
@@ -930,7 +952,7 @@ bt_target_page_check(BtreeCheckState *state)
* and probably not markedly more effective in practice.
*/
if (!P_RIGHTMOST(topaque) &&
- !invariant_leq_offset(state, skey, P_HIKEY))+ !invariant_leq_offset(state, tupnkeyatts, skey, scantid, P_HIKEY))
{
char *itid,
*htid;
@@ -956,11 +978,11 @@ bt_target_page_check(BtreeCheckState *state)
* * Item order check *
*
* Check that items are stored on page in logical order, by checking
- * current item is less than or equal to next item (if any).+ * current item is strictly less than next item (if any).
*/
if (OffsetNumberNext(offset) <= max &&
- !invariant_leq_offset(state, skey,- OffsetNumberNext(offset)))+ !invariant_l_offset(state, tupnkeyatts, skey, scantid,+ OffsetNumberNext(offset)))
{
char *itid,
*htid,
@@ -1017,16 +1039,29 @@ bt_target_page_check(BtreeCheckState *state)
*/
else if (offset == max)
{
+ IndexTuple righttup;
ScanKey rightkey;
+ int righttupnkeyatts;+ ItemPointer rightscantid;
/* Get item in next/right page */
- rightkey = bt_right_page_check_scankey(state);+ righttup = bt_right_page_check_tuple(state);++ /* Set up right item scankey */+ if (righttup)+ {+ righttupnkeyatts = BTreeTupleGetNKeyAtts(righttup, state->rel);+ rightkey = _bt_mkscankey(state->rel, righttup);+ rightscantid = BTreeTupleGetHeapTIDCareful(state, righttup,+ P_ISLEAF(topaque));+ }- if (rightkey &&- !invariant_geq_offset(state, rightkey, max))+ if (righttup &&+ !invariant_g_offset(state, righttupnkeyatts, rightkey,+ rightscantid, max))
{
/*
- * As explained at length in bt_right_page_check_scankey(),+ * As explained at length in bt_right_page_check_tuple(),
* there is a known !readonly race that could account for
* apparent violation of invariant, which we must check for
* before actually proceeding with raising error. Our canary
@@ -1069,7 +1104,7 @@ bt_target_page_check(BtreeCheckState *state)
{
BlockNumber childblock = BTreeInnerTupleGetDownLink(itup);
- bt_downlink_check(state, childblock, skey);+ bt_downlink_check(state, childblock, skey, scantid, tupnkeyatts);
}
}
@@ -1083,9 +1118,9 @@ bt_target_page_check(BtreeCheckState *state)
}
/*
- * Return a scankey for an item on page to right of current target (or the+ * Return an index tuple for an item on page to right of current target (or the
* first non-ignorable page), sufficient to check ordering invariant on last
- * item in current target page. Returned scankey relies on local memory+ * item in current target page. Returned tuple relies on local memory
* allocated for the child page, which caller cannot pfree(). Caller's memory
* context should be reset between calls here.
*
@@ -1098,8 +1133,8 @@ bt_target_page_check(BtreeCheckState *state)
* Note that !readonly callers must reverify that target page has not
* been concurrently deleted.
*/
-static ScanKey-bt_right_page_check_scankey(BtreeCheckState *state)+static IndexTuple+bt_right_page_check_tuple(BtreeCheckState *state)
{
BTPageOpaque opaque;
ItemId rightitem;
@@ -1287,11 +1322,10 @@ bt_right_page_check_scankey(BtreeCheckState *state)
}
/*
- * Return first real item scankey. Note that this relies on right page- * memory remaining allocated.+ * Return first real item. Note that this relies on right page memory+ * remaining allocated.
*/
- return _bt_mkscankey(state->rel,- (IndexTuple) PageGetItem(rightpage, rightitem));+ return (IndexTuple) PageGetItem(rightpage, rightitem);
}
/*
@@ -1305,7 +1339,7 @@ bt_right_page_check_scankey(BtreeCheckState *state)
*/
static void
bt_downlink_check(BtreeCheckState *state, BlockNumber childblock,
- ScanKey targetkey)+ ScanKey targetkey, ItemPointer scantid, int tupnkeyatts)
{
OffsetNumber offset;
OffsetNumber maxoffset;
@@ -1354,7 +1388,7 @@ bt_downlink_check(BtreeCheckState *state, BlockNumber childblock,
/*
* Verify child page has the downlink key from target page (its parent) as
- * a lower bound.+ * a lower bound; downlink must be strictly less than all keys on the page.
*
* Check all items, rather than checking just the first and trusting that
* the operator class obeys the transitive law.
@@ -1404,14 +1438,14 @@ bt_downlink_check(BtreeCheckState *state, BlockNumber childblock,
/*
* Skip comparison of target page key against "negative infinity"
* item, if any. Checking it would indicate that it's not an upper
- * bound, but that's only because of the hard-coding within- * _bt_compare().+ * bound, but that's only because of the hard-coding for negative+ * inifinity items within _bt_compare().
*/
if (offset_is_negative_infinity(copaque, offset))
continue;
- if (!invariant_leq_nontarget_offset(state, child,- targetkey, offset))+ if (!invariant_l_nontarget_offset(state, child, tupnkeyatts, targetkey,+ scantid, offset))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg("down-link lower bound invariant violated for index \"%s\"",
@@ -1751,6 +1785,54 @@ offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
return !P_ISLEAF(opaque) && offset == P_FIRSTDATAKEY(opaque);
}
+/*+ * Does the invariant hold that the key is strictly less than a given upper+ * bound offset item?+ *+ * If this function returns false, convention is that caller throws error due+ * to corruption.+ */+static inline bool+invariant_l_offset(BtreeCheckState *state, int tupnkeyatts, ScanKey key,+ ItemPointer scantid, OffsetNumber upperbound)+{+ int32 cmp;++ cmp = _bt_compare(state->rel, tupnkeyatts, key, scantid, state->target,+ upperbound);++ /*+ * _bt_compare interprets the absence of attributes in scan keys as meaning+ * that they're not participating in a search, not as negative infinity+ * (only tuples within the index are treated as negative infinity).+ * Compensate for that here.+ */+ if (cmp == 0)+ {+ BTPageOpaque topaque;+ ItemId itemid;+ IndexTuple ritup;+ int uppnkeyatts;+ ItemPointer rheaptid;++ itemid = PageGetItemId(state->target, upperbound);+ ritup = (IndexTuple) PageGetItem(state->target, itemid);+ uppnkeyatts = BTreeTupleGetNKeyAtts(ritup, state->rel);++ /* Get heap TID for item to the right */+ topaque = (BTPageOpaque) PageGetSpecialPointer(state->target);+ rheaptid = BTreeTupleGetHeapTIDCareful(state, ritup,+ P_ISLEAF(topaque));++ if (uppnkeyatts == tupnkeyatts)+ return scantid == NULL && rheaptid != NULL;++ return tupnkeyatts < uppnkeyatts;+ }++ return cmp < 0;+}+
/*
* Does the invariant hold that the key is less than or equal to a given upper
* bound offset item?
@@ -1759,57 +1841,93 @@ offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
* to corruption.
*/
static inline bool
-invariant_leq_offset(BtreeCheckState *state, ScanKey key,- OffsetNumber upperbound)+invariant_leq_offset(BtreeCheckState *state, int tupnkeyatts, ScanKey key,+ ItemPointer scantid, OffsetNumber upperbound)
{
- int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(state->rel);
int32 cmp;
- cmp = _bt_compare(state->rel, nkeyatts, key, state->target, upperbound);+ cmp = _bt_compare(state->rel, tupnkeyatts, key, scantid, state->target,+ upperbound);
return cmp <= 0;
}
/*
- * Does the invariant hold that the key is greater than or equal to a given- * lower bound offset item?+ * Does the invariant hold that the key is strictly greater than a given lower+ * bound offset item?
*
* If this function returns false, convention is that caller throws error due
* to corruption.
*/
static inline bool
-invariant_geq_offset(BtreeCheckState *state, ScanKey key,- OffsetNumber lowerbound)+invariant_g_offset(BtreeCheckState *state, int tupnkeyatts, ScanKey key,+ ItemPointer scantid, OffsetNumber lowerbound)
{
- int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(state->rel);
int32 cmp;
- cmp = _bt_compare(state->rel, nkeyatts, key, state->target, lowerbound);+ /*+ * No need to consider possibility that scankey has attributes that we need+ * to force to be interpreted as negative infinity, since scan key has to+ * be strictly greater than lower bound offset.+ */+ cmp = _bt_compare(state->rel, tupnkeyatts, key, scantid, state->target,+ lowerbound);- return cmp >= 0;+ return cmp > 0;
}
/*
- * Does the invariant hold that the key is less than or equal to a given upper+ * Does the invariant hold that the key is strictly less than a given upper
* bound offset item, with the offset relating to a caller-supplied page that
- * is not the current target page? Caller's non-target page is typically a- * child page of the target, checked as part of checking a property of the- * target page (i.e. the key comes from the target).+ * is not the current target page?+ *+ * Caller's non-target page is a child page of the target, checked as part of+ * checking a property of the target page (i.e. the key comes from the+ * target).
*
* If this function returns false, convention is that caller throws error due
* to corruption.
*/
static inline bool
-invariant_leq_nontarget_offset(BtreeCheckState *state,- Page nontarget, ScanKey key,- OffsetNumber upperbound)+invariant_l_nontarget_offset(BtreeCheckState *state, Page nontarget,+ int tupnkeyatts, ScanKey key, ItemPointer scantid,+ OffsetNumber upperbound)
{
- int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(state->rel);
int32 cmp;
- cmp = _bt_compare(state->rel, nkeyatts, key, nontarget, upperbound);+ cmp = _bt_compare(state->rel, tupnkeyatts, key, scantid, nontarget,+ upperbound);- return cmp <= 0;+ /*+ * _bt_compare interprets the absence of attributes in scan keys as meaning+ * that they're not participating in a search, not as negative infinity+ * (only tuples within the index are treated as negative infinity).+ * Compensate for that here.+ */+ if (cmp == 0)+ {+ ItemId itemid;+ IndexTuple child;+ int uppnkeyatts;+ ItemPointer childheaptid;+ BTPageOpaque copaque;++ copaque = (BTPageOpaque) PageGetSpecialPointer(nontarget);+ itemid = PageGetItemId(nontarget, upperbound);+ child = (IndexTuple) PageGetItem(nontarget, itemid);+ uppnkeyatts = BTreeTupleGetNKeyAtts(child, state->rel);++ /* Get heap TID for item from child/non-target */+ childheaptid = BTreeTupleGetHeapTIDCareful(state, child,+ P_ISLEAF(copaque));++ if (uppnkeyatts == tupnkeyatts)+ return scantid == NULL && childheaptid != NULL;++ return tupnkeyatts < uppnkeyatts;+ }++ return cmp < 0;
}
/*
@@ -1965,3 +2083,31 @@ palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum)
return page;
}
++/*+ * BTreeTupleGetHeapTID() wrapper that lets caller enforce that a heap TID must+ * be present in cases where that is mandatory.+ *+ * This doesn't add much as of BTREE_VERSION 4, since the INDEX_ALT_TID_MASK+ * bit is effectively a proxy for whether or not the tuple is a pivot tuple.+ * It may become more useful in the future, when non-pivot tuples support their+ * own alternative INDEX_ALT_TID_MASK representation.+ *+ * Note that it is incorrect to specify the tuple as a non-pivot when passing a+ * leaf tuple that came from the high key offset, since that is actually a+ * pivot tuple.+ */+static inline ItemPointer+BTreeTupleGetHeapTIDCareful(BtreeCheckState *state, IndexTuple itup,+ bool nonpivot)+{+ ItemPointer result = BTreeTupleGetHeapTID(itup);++ if (result == NULL && nonpivot)+ ereport(ERROR,+ (errcode(ERRCODE_INDEX_CORRUPTED),+ errmsg("block %u or its right sibling block or child block in index \"%s\" contains non-pivot tuple that lacks a heap TID",+ state->targetblock, RelationGetRelationName(state->rel))));++ return result;+}diff --git a/contrib/pageinspect/expected/btree.out b/contrib/pageinspect/expected/btree.outindex 2aaa4df53b..07c2dcd771 100644--- a/contrib/pageinspect/expected/btree.out+++ b/contrib/pageinspect/expected/btree.out@@ -5,7 +5,7 @@ CREATE INDEX test1_a_idx ON test1 USING btree (a);
SELECT * FROM bt_metap('test1_a_idx');
-[ RECORD 1 ]-----------+-------
magic | 340322
-version | 3+version | 4
root | 1
level | 0
fastroot | 1
diff --git a/contrib/pgstattuple/expected/pgstattuple.out b/contrib/pgstattuple/expected/pgstattuple.outindex 9858ea69d4..9920dbfd40 100644--- a/contrib/pgstattuple/expected/pgstattuple.out+++ b/contrib/pgstattuple/expected/pgstattuple.out@@ -48,7 +48,7 @@ select version, tree_level,
from pgstatindex('test_pkey');
version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
- 3 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN+ 4 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN
(1 row)
select version, tree_level,
@@ -58,7 +58,7 @@ select version, tree_level,
from pgstatindex('test_pkey'::text);
version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
- 3 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN+ 4 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN
(1 row)
select version, tree_level,
@@ -68,7 +68,7 @@ select version, tree_level,
from pgstatindex('test_pkey'::name);
version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
- 3 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN+ 4 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN
(1 row)
select version, tree_level,
@@ -78,7 +78,7 @@ select version, tree_level,
from pgstatindex('test_pkey'::regclass);
version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
- 3 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN+ 4 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | NaN | NaN
(1 row)
select pg_relpages('test');
@@ -232,7 +232,7 @@ create index test_partition_hash_idx on test_partition using hash (a);
select pgstatindex('test_partition_idx');
pgstatindex
------------------------------
- (3,0,8192,0,0,0,0,0,NaN,NaN)+ (4,0,8192,0,0,0,0,0,NaN,NaN)
(1 row)
select pgstathashindex('test_partition_hash_idx');
diff --git a/src/backend/access/nbtree/README b/src/backend/access/nbtree/READMEindex 3680e69b89..dc6c65d201 100644--- a/src/backend/access/nbtree/README+++ b/src/backend/access/nbtree/README@@ -34,30 +34,47 @@ Differences to the Lehman & Yao algorithm
We have made the following changes in order to incorporate the L&Y algorithm
into Postgres:
-The requirement that all btree keys be unique is too onerous,-but the algorithm won't work correctly without it. Fortunately, it is-only necessary that keys be unique on a single tree level, because L&Y-only use the assumption of key uniqueness when re-finding a key in a-parent page (to determine where to insert the key for a split page).-Therefore, we can use the link field to disambiguate multiple-occurrences of the same user key: only one entry in the parent level-will be pointing at the page we had split. (Indeed we need not look at-the real "key" at all, just at the link field.) We can distinguish-items at the leaf level in the same way, by examining their links to-heap tuples; we'd never have two items for the same heap tuple.--Lehman and Yao assume that the key range for a subtree S is described-by Ki < v <= Ki+1 where Ki and Ki+1 are the adjacent keys in the parent-page. This does not work for nonunique keys (for example, if we have-enough equal keys to spread across several leaf pages, there *must* be-some equal bounding keys in the first level up). Therefore we assume-Ki <= v <= Ki+1 instead. A search that finds exact equality to a-bounding key in an upper tree level must descend to the left of that-key to ensure it finds any equal keys in the preceding page. An-insertion that sees the high key of its target page is equal to the key-to be inserted has a choice whether or not to move right, since the new-key could go on either page. (Currently, we try to find a page where-there is room for the new key without a split.)+The requirement that all btree keys be unique is satisfied by treating+heap TID as a tie-breaker attribute. Logical duplicates are sorted in+descending item pointer order. We don't use btree keys to+disambiguate downlinks from the internal pages during a page split,+though: only one entry in the parent level will be pointing at the+page we just split, so the link fields can be used to re-find+downlinks in the parent via a linear search.++Lehman and Yao require that the key range for a subtree S is described+by Ki < v <= Ki+1 where Ki and Ki+1 are the adjacent keys in the+parent page, but do not account for the need to search the tree based+only on leading index attributes in a composite index. Since heap TID+is always used to make btree keys unique (even in unique indexes),+every btree index is treated as a composite index internally. A+search that finds exact equality to a pivot tuple in an upper tree+level must descend to the left of that key to ensure it finds any+equal keys, even when scan values were provided for all attributes.+An insertion that sees that the high key of its target page is equal+to the key to be inserted cannot move right, since the downlink for+the right sibling in the parent must always be strictly less than+right sibling keys (this is always possible because the leftmost+downlink on any non-leaf level is always a negative infinity+downlink).++We might be able to avoid moving left in the event of a full match on+all attributes up to and including the heap TID attribute, but that+would be a very narrow win, since it's rather unlikely that heap TID+will be an exact match. We can avoid moving left unnecessarily when+all user-visible keys are equal by avoiding exact equality; a+sentinel value that's less than any possible heap TID is used by most+index scans. This is effective because of suffix truncation. An+"extra" heap TID attribute in pivot tuples is almost always avoided.+All truncated attributes compare as minus infinity, even against a+sentinel value, and the sentinel value is less than any real TID+value, so an unnecessary move to the left is avoided regardless of+whether or not a heap TID is present in the otherwise-equal pivot+tuple. Consistently moving left on full equality is also needed by+page deletion, which re-finds a leaf page by descending the tree while+searching on the leaf page's high key. If we wanted to avoid moving+left without breaking page deletion, we'd have to avoid suffix+truncation, which could never be worth it.
Lehman and Yao don't require read locks, but assume that in-memory
copies of tree pages are unshared. Postgres shares in-memory buffers
@@ -610,21 +627,25 @@ scanned to decide whether to return the entry and whether the scan can
stop (see _bt_checkkeys()).
We use term "pivot" index tuples to distinguish tuples which don't point
-to heap tuples, but rather used for tree navigation. Pivot tuples includes-all tuples on non-leaf pages and high keys on leaf pages. Note that pivot-index tuples are only used to represent which part of the key space belongs-on each page, and can have attribute values copied from non-pivot tuples-that were deleted and killed by VACUUM some time ago. In principle, we could-truncate away attributes that are not needed for a page high key during a leaf-page split, provided that the remaining attributes distinguish the last index-tuple on the post-split left page as belonging on the left page, and the first-index tuple on the post-split right page as belonging on the right page. This-optimization is sometimes called suffix truncation, and may appear in a future-release. Since the high key is subsequently reused as the downlink in the-parent page for the new right page, suffix truncation can increase index-fan-out considerably by keeping pivot tuples short. INCLUDE indexes similarly-truncate away non-key attributes at the time of a leaf page split,-increasing fan-out.+to heap tuples, that are used only for tree navigation. Pivot tuples+includes all tuples on non-leaf pages and high keys on leaf pages. Note+that pivot index tuples are only used to represent which part of the key+space belongs on each page, and can have attribute values copied from+non-pivot tuples that were deleted and killed by VACUUM some time ago.++We truncate away attributes that are not needed for a page high key during+a leaf page split, provided that the remaining attributes distinguish the+last index tuple on the post-split left page as belonging on the left+page, and the first index tuple on the post-split right page as belonging+on the right page. A truncated tuple logically retains the truncated+suffix key attributes, which implicitly have "negative infinity" as their+value. This optimization is called suffix truncation. Since the high key+is subsequently reused as the downlink in the parent page for the new+right page, suffix truncation can increase index fan-out considerably by+keeping pivot tuples short. INCLUDE indexes are guaranteed to have+non-key attributes truncated at the time of a leaf page split, but may+also have some key attributes truncated away, based on the usual criteria+for key attributes.
Notes About Data Representation
-------------------------------
@@ -658,4 +679,19 @@ downlink. The first data item on each such page has no lower bound
routines must treat it accordingly. The actual key stored in the
item is irrelevant, and need not be stored at all. This arrangement
corresponds to the fact that an L&Y non-leaf page has one more pointer
-than key.+than key. Suffix truncation's negative infinity attributes behave in+the same way.++Non-leaf pages only truly need to truncate their first item to zero+attributes at the leftmost level, since that truly is negative infinity.+All other negative infinity items are only really negative infinity+within the subtree that the page is at the root of (or is a leftmost+page within). We truncate away all attributes of the first item on+non-leaf pages just the same, to save a little space. If we ever+avoided zero-truncating items on pages where that doesn't accurately+represent the absolute separation of the keyspace, we'd be left with+"low key" items on internal pages -- a key value that can be used as a+lower bound on items on the page, much like the high key is an upper+bound. (Actually, that would even be true of "true" negative infinity+items. One can think of rightmost pages as implicitly containing+"positive infinity" high keys.)diff --git a/src/backend/access/nbtree/nbtinsert.c b/src/backend/access/nbtree/nbtinsert.cindex 4b2b4746f7..7d0556b91d 100644--- a/src/backend/access/nbtree/nbtinsert.c+++ b/src/backend/access/nbtree/nbtinsert.c@@ -76,9 +76,11 @@ static Buffer _bt_split(Relation rel, Buffer buf, Buffer cbuf,
static void _bt_insert_parent(Relation rel, Buffer buf, Buffer rbuf,
BTStack stack, bool is_root, bool is_only);
static OffsetNumber _bt_findsplitloc(Relation rel, Page page,
+ bool disjunctpass,
OffsetNumber newitemoff,
Size newitemsz,
bool *newitemonleft);
+static bool _bt_split_isdisjunct(Page page, OffsetNumber item);
static void _bt_checksplitloc(FindSplitData *state,
OffsetNumber firstoldonright, bool newitemonleft,
int dataitemstoleft, Size firstoldonrightsz);
@@ -113,9 +115,12 @@ _bt_doinsert(Relation rel, IndexTuple itup,
bool is_unique = false;
int indnkeyatts;
ScanKey itup_scankey;
+ ItemPointer itup_scantid;
BTStack stack = NULL;
Buffer buf;
OffsetNumber offset;
+ Page page;+ BTPageOpaque lpageop;
bool fastpath;
indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
@@ -123,6 +128,8 @@ _bt_doinsert(Relation rel, IndexTuple itup,
/* we need an insertion scan key to do our search, so build one */
itup_scankey = _bt_mkscankey(rel, itup);
+ /* we use a heap TID with scan key if this isn't unique case */+ itup_scantid = (checkUnique == UNIQUE_CHECK_NO ? &itup->t_tid : NULL);
/*
* It's very common to have an index on an auto-incremented or
@@ -149,8 +156,6 @@ top:
if (RelationGetTargetBlock(rel) != InvalidBlockNumber)
{
Size itemsz;
- Page page;- BTPageOpaque lpageop;
/*
* Conditionally acquire exclusive lock on the buffer before doing any
@@ -180,7 +185,7 @@ top:
!P_IGNORE(lpageop) &&
(PageGetFreeSpace(page) > itemsz) &&
PageGetMaxOffsetNumber(page) >= P_FIRSTDATAKEY(lpageop) &&
- _bt_compare(rel, indnkeyatts, itup_scankey, page,+ _bt_compare(rel, indnkeyatts, itup_scankey, itup_scantid, page,
P_FIRSTDATAKEY(lpageop)) > 0)
{
/*
@@ -217,8 +222,8 @@ top:
if (!fastpath)
{
/* find the first page containing this key */
- stack = _bt_search(rel, indnkeyatts, itup_scankey, false, &buf, BT_WRITE,- NULL);+ stack = _bt_search(rel, indnkeyatts, itup_scankey, itup_scantid, false,+ &buf, BT_WRITE, NULL);
/* trade in our read lock for a write lock */
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
@@ -231,8 +236,8 @@ top:
* need to move right in the tree. See Lehman and Yao for an
* excruciatingly precise description.
*/
- buf = _bt_moveright(rel, buf, indnkeyatts, itup_scankey, false,- true, stack, BT_WRITE, NULL);+ buf = _bt_moveright(rel, buf, indnkeyatts, itup_scankey, itup_scantid,+ false, true, stack, BT_WRITE, NULL);
}
/*
@@ -242,12 +247,13 @@ top:
* NOTE: obviously, _bt_check_unique can only detect keys that are already
* in the index; so it cannot defend against concurrent insertions of the
* same key. We protect against that by means of holding a write lock on
- * the target page. Any other would-be inserter of the same key must- * acquire a write lock on the same target page, so only one would-be- * inserter can be making the check at one time. Furthermore, once we are- * past the check we hold write locks continuously until we have performed- * our insertion, so no later inserter can fail to see our insertion.- * (This requires some care in _bt_findinsertloc.)+ * the first page the value could be on, regardless of the value of its+ * implicit heap TID tie-breaker attribute. Any other would-be inserter of+ * the same key must acquire a write lock on the same page, so only one+ * would-be inserter can be making the check at one time. Furthermore,+ * once we are past the check we hold write locks continuously until we+ * have performed our insertion, so no later inserter can fail to see our+ * insertion. (This requires some care in _bt_findinsertloc.)
*
* If we must wait for another xact, we release the lock while waiting,
* and then must start over completely.
@@ -261,7 +267,11 @@ top:
TransactionId xwait;
uint32 speculativeToken;
- offset = _bt_binsrch(rel, buf, indnkeyatts, itup_scankey, false);+ page = BufferGetPage(buf);+ lpageop = (BTPageOpaque) PageGetSpecialPointer(page);+ Assert(itup_scantid == NULL);+ offset = _bt_binsrch(rel, buf, indnkeyatts, itup_scankey, NULL,+ P_FIRSTDATAKEY(lpageop), false);
xwait = _bt_check_unique(rel, itup, heapRel, buf, offset, itup_scankey,
checkUnique, &is_unique, &speculativeToken);
@@ -299,7 +309,7 @@ top:
* attributes are not considered part of the key space.
*/
CheckForSerializableConflictIn(rel, NULL, buf);
- /* do the insertion */+ /* do the insertion, possibly on a page to the right in unique case */
_bt_findinsertloc(rel, &buf, &offset, indnkeyatts, itup_scankey, itup,
stack, heapRel);
_bt_insertonpg(rel, buf, InvalidBuffer, stack, itup, offset, false);
@@ -564,11 +574,11 @@ _bt_check_unique(Relation rel, IndexTuple itup, Relation heapRel,
offset = OffsetNumberNext(offset);
else
{
- /* If scankey == hikey we gotta check the next page too */+ /* If scankey <= hikey we gotta check the next page too */
if (P_RIGHTMOST(opaque))
break;
- if (!_bt_isequal(itupdesc, page, P_HIKEY,- indnkeyatts, itup_scankey))+ /* _bt_isequal()'s special NULL semantics not required here */+ if (_bt_compare(rel, indnkeyatts, itup_scankey, NULL, page, P_HIKEY) > 0)
break;
/* Advance to next non-dead page --- there must be one */
for (;;)
@@ -612,31 +622,22 @@ _bt_check_unique(Relation rel, IndexTuple itup, Relation heapRel,
/*
* _bt_findinsertloc() -- Finds an insert location for a tuple
*
- * If the new key is equal to one or more existing keys, we can- * legitimately place it anywhere in the series of equal keys --- in fact,- * if the new key is equal to the page's "high key" we can place it on- * the next page. If it is equal to the high key, and there's not room- * to insert the new tuple on the current page without splitting, then- * we can move right hoping to find more free space and avoid a split.- * (We should not move right indefinitely, however, since that leads to- * O(N^2) insertion behavior in the presence of many equal keys.)- * Once we have chosen the page to put the key on, we'll insert it before- * any existing equal keys because of the way _bt_binsrch() works.- *- * If there's not enough room in the space, we try to make room by- * removing any LP_DEAD tuples.- *
* On entry, *bufptr and *offsetptr point to the first legal position
- * where the new tuple could be inserted. The caller should hold an- * exclusive lock on *bufptr. *offsetptr can also be set to- * InvalidOffsetNumber, in which case the function will search for the- * right location within the page if needed. On exit, they point to the- * chosen insert location. If _bt_findinsertloc decides to move right,- * the lock and pin on the original page will be released and the new- * page returned to the caller is exclusively locked instead.+ * where the new tuple could be inserted if we were to treat it as having+ * no implicit heap TID; only callers that just called _bt_check_unique()+ * provide this hint (all other callers should set *offsetptr to+ * InvalidOffsetNumber). The caller should hold an exclusive lock on+ * *bufptr in all cases. On exit, they both point to the chosen insert+ * location in all cases. If _bt_findinsertloc decides to move right, the+ * lock and pin on the original page will be released, and the new page+ * returned to the caller is exclusively locked instead.+ *+ * This is also where opportunistic microvacuuming of LP_DEAD tuples+ * occurs.
*
* newtup is the new tuple we're inserting, and scankey is an insertion
- * type scan key for it.+ * type scan key for it. We take a "scantid" heap TID attribute value+ * from newtup directly.
*/
static void
_bt_findinsertloc(Relation rel,
@@ -652,9 +653,9 @@ _bt_findinsertloc(Relation rel,
Page page = BufferGetPage(buf);
Size itemsz;
BTPageOpaque lpageop;
- bool movedright,- vacuumed;+ bool hintinvalidated;
OffsetNumber newitemoff;
+ OffsetNumber lowitemoff;
OffsetNumber firstlegaloff = *offsetptr;
lpageop = (BTPageOpaque) PageGetSpecialPointer(page);
@@ -684,59 +685,30 @@ _bt_findinsertloc(Relation rel,
errtableconstraint(heapRel,
RelationGetRelationName(rel))));
- /*----------- * If we will need to split the page to put the item on this page,- * check whether we can put the tuple somewhere to the right,- * instead. Keep scanning right until we- * (a) find a page with enough free space,- * (b) reach the last page where the tuple can legally go, or- * (c) get tired of searching.- * (c) is not flippant; it is important because if there are many- * pages' worth of equal keys, it's better to split one of the early- * pages than to scan all the way to the end of the run of equal keys- * on every insert. We implement "get tired" as a random choice,- * since stopping after scanning a fixed number of pages wouldn't work- * well (we'd never reach the right-hand side of previously split- * pages). Currently the probability of moving right is set at 0.99,- * which may seem too high to change the behavior much, but it does an- * excellent job of preventing O(N^2) behavior with many equal keys.- *----------+ /* firstlegaloff/offsetptr hint (if any) assumed valid initially */+ hintinvalidated = false;++ /*+ * TODO: Restore the logic for finding a page to insert on in the event of+ * many duplicates for pre-pg_upgrade indexes. The whole search through+ * pages of logical duplicates to determine where to insert seems like+ * something that has little upside, but that doesn't make it okay to+ * ignore the performance characteristics after pg_upgrade is run, but+ * before a REINDEX can run to bump BTREE_VERSION.
*/
- movedright = false;- vacuumed = false;- while (PageGetFreeSpace(page) < itemsz)+ while (true)
{
Buffer rbuf;
BlockNumber rblkno;
- /*- * before considering moving right, see if we can obtain enough space- * by erasing LP_DEAD items- */- if (P_ISLEAF(lpageop) && P_HAS_GARBAGE(lpageop))- {- _bt_vacuum_one_page(rel, buf, heapRel);-- /*- * remember that we vacuumed this page, because that makes the- * hint supplied by the caller invalid- */- vacuumed = true;-- if (PageGetFreeSpace(page) >= itemsz)- break; /* OK, now we have enough space */- }-- /*- * nope, so check conditions (b) and (c) enumerated above- */
if (P_RIGHTMOST(lpageop) ||
- _bt_compare(rel, keysz, scankey, page, P_HIKEY) != 0 ||- random() <= (MAX_RANDOM_VALUE / 100))+ _bt_compare(rel, keysz, scankey, &newtup->t_tid, page, P_HIKEY) <= 0)
break;
/*
- * step right to next non-dead page+ * step right to next non-dead page. this is only needed for unique+ * indexes, and pg_upgrade'd indexes that still use BTREE_VERSION 2 or+ * 3, where heap TID isn't considered to be a part of the keyspace.
*
* must write-lock that page before releasing write lock on current
* page; else someone else's _bt_check_unique scan could fail to see
@@ -775,24 +747,40 @@ _bt_findinsertloc(Relation rel,
}
_bt_relbuf(rel, buf);
buf = rbuf;
- movedright = true;- vacuumed = false;+ hintinvalidated = true;
}
+ Assert(P_ISLEAF(lpageop));+
/*
- * Now we are on the right page, so find the insert position. If we moved- * right at all, we know we should insert at the start of the page. If we- * didn't move right, we can use the firstlegaloff hint if the caller- * supplied one, unless we vacuumed the page which might have moved tuples- * around making the hint invalid. If we didn't move right or can't use- * the hint, find the position by searching.+ * Perform micro-vacuuming of the page we're about to insert tuple on to if+ * it looks like it has LP_DEAD items.
*/
- if (movedright)- newitemoff = P_FIRSTDATAKEY(lpageop);- else if (firstlegaloff != InvalidOffsetNumber && !vacuumed)- newitemoff = firstlegaloff;+ if (P_HAS_GARBAGE(lpageop) && PageGetFreeSpace(page) < itemsz)+ {+ _bt_vacuum_one_page(rel, buf, heapRel);++ hintinvalidated = true;+ }++ /*+ * Consider using caller's hint to avoid repeated binary search effort.+ *+ * Note that the hint is only provided by callers that checked uniqueness.+ * The hint is used as a lower bound for a new binary search, since+ * caller's original binary search won't have specified a scan tid.+ */+ if (firstlegaloff == InvalidOffsetNumber || hintinvalidated)+ lowitemoff = P_FIRSTDATAKEY(lpageop);
else
- newitemoff = _bt_binsrch(rel, buf, keysz, scankey, false);+ {+ Assert(firstlegaloff == _bt_binsrch(rel, buf, keysz, scankey, NULL,+ P_FIRSTDATAKEY(lpageop), false));+ lowitemoff = firstlegaloff;+ }++ newitemoff = _bt_binsrch(rel, buf, keysz, scankey, &newtup->t_tid,+ lowitemoff, false);
*bufptr = buf;
*offsetptr = newitemoff;
@@ -851,11 +839,12 @@ _bt_insertonpg(Relation rel,
/* child buffer must be given iff inserting on an internal page */
Assert(P_ISLEAF(lpageop) == !BufferIsValid(cbuf));
/* tuple must have appropriate number of attributes */
+ Assert(BTreeTupleGetNAtts(itup, rel) > 0);
Assert(!P_ISLEAF(lpageop) ||
BTreeTupleGetNAtts(itup, rel) ==
IndexRelationGetNumberOfAttributes(rel));
Assert(P_ISLEAF(lpageop) ||
- BTreeTupleGetNAtts(itup, rel) ==+ BTreeTupleGetNAtts(itup, rel) <=
IndexRelationGetNumberOfKeyAttributes(rel));
/* The caller should've finished any incomplete splits already. */
@@ -900,7 +889,7 @@ _bt_insertonpg(Relation rel,
BlockNumberIsValid(RelationGetTargetBlock(rel))));
/* Choose the split point */
- firstright = _bt_findsplitloc(rel, page,+ firstright = _bt_findsplitloc(rel, page, true,
newitemoff, itemsz,
&newitemonleft);
@@ -1143,8 +1132,6 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
OffsetNumber i;
bool isleaf;
IndexTuple lefthikey;
- int indnatts = IndexRelationGetNumberOfAttributes(rel);- int indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
/* Acquire a new page to split into */
rbuf = _bt_getbuf(rel, P_NEW, BT_WRITE);
@@ -1214,7 +1201,9 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
itemid = PageGetItemId(origpage, P_HIKEY);
itemsz = ItemIdGetLength(itemid);
item = (IndexTuple) PageGetItem(origpage, itemid);
- Assert(BTreeTupleGetNAtts(item, rel) == indnkeyatts);+ Assert(BTreeTupleGetNAtts(item, rel) > 0);+ Assert(BTreeTupleGetNAtts(item, rel) <=+ IndexRelationGetNumberOfKeyAttributes(rel));
if (PageAddItem(rightpage, (Item) item, itemsz, rightoff,
false, false) == InvalidOffsetNumber)
{
@@ -1228,8 +1217,9 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
/*
* The "high key" for the new left page will be the first key that's going
- * to go into the new right page. This might be either the existing data- * item at position firstright, or the incoming tuple.+ * to go into the new right page, or possibly a truncated version if this+ * is a leaf page split. This might be either the existing data item at+ * position firstright, or the incoming tuple.
*/
leftoff = P_HIKEY;
if (!newitemonleft && newitemoff == firstright)
@@ -1247,25 +1237,77 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
}
/*
- * Truncate non-key (INCLUDE) attributes of the high key item before- * inserting it on the left page. This only needs to happen at the leaf- * level, since in general all pivot tuple values originate from leaf- * level high keys. This isn't just about avoiding unnecessary work,- * though; truncating unneeded key attributes (more aggressive suffix- * truncation) can only be performed at the leaf level anyway. This is- * because a pivot tuple in a grandparent page must guide a search not- * only to the correct parent page, but also to the correct leaf page.+ * Truncate attributes of the high key item before inserting it on the left+ * page. This can only happen at the leaf level, since in general all+ * pivot tuple values originate from leaf level high keys. This isn't just+ * about avoiding unnecessary work, though; truncating unneeded key suffix+ * attributes can only be performed at the leaf level anyway. This is+ * because a pivot tuple in a grandparent page must guide a search not only+ * to the correct parent page, but also to the correct leaf page.+ *+ * Note that non-key (INCLUDE) attributes are always truncated away here.+ * Additional key attributes are truncated away when they're not required+ * to correctly separate the key space.
*/
- if (indnatts != indnkeyatts && isleaf)+ if (isleaf)
{
- lefthikey = _bt_nonkey_truncate(rel, item);+ OffsetNumber lastleftoff;+ IndexTuple lastleft;++ /*+ * Determine which tuple is on the left side of the split point, and+ * generate truncated copy of the right tuple. Truncate as+ * aggressively as possible without generating a high key for the left+ * side of the split (and later downlink for the right side) that fails+ * to distinguish each side. The new high key needs to be strictly+ * less than all tuples on the right side of the split, but can be+ * equal to items on the left side of the split. We almost always find+ * a way to make it strictly greater than lastleft, though.+ *+ * Handle the case where the incoming tuple is about to become the last+ * item on the left side of the split.+ */+ if (newitemonleft && newitemoff == firstright)+ lastleft = newitem;+ else+ {+ lastleftoff = OffsetNumberPrev(firstright);+ itemid = PageGetItemId(origpage, lastleftoff);+ lastleft = (IndexTuple) PageGetItem(origpage, itemid);+ }++ Assert(lastleft != item);+ lefthikey = _bt_suffix_truncate(rel, lastleft, item);
itemsz = IndexTupleSize(lefthikey);
itemsz = MAXALIGN(itemsz);
+#ifdef DEBUG_SPLITS+ {+ TupleDesc itupdesc = RelationGetDescr(rel);+ Datum values[INDEX_MAX_KEYS];+ bool isnull[INDEX_MAX_KEYS];+ char *lastleftstr;+ char *firstrightstr;++ index_deform_tuple(lastleft, itupdesc, values, isnull);+ lastleftstr = BuildIndexValueDescription(rel, values, isnull);+ index_deform_tuple(item, itupdesc, values, isnull);+ firstrightstr = BuildIndexValueDescription(rel, values, isnull);++ elog(LOG, "split of block %u "+ "last left %s first right %s "+ "attributes in new high key for left page %u%s",+ BufferGetBlockNumber(buf), lastleftstr, firstrightstr,+ BTreeTupleGetNAtts(lefthikey, rel),+ BTreeTupleGetHeapTID(lefthikey) != NULL ? " (plus heap TID)":"");+ }+#endif
}
else
lefthikey = item;
- Assert(BTreeTupleGetNAtts(lefthikey, rel) == indnkeyatts);+ Assert(BTreeTupleGetNAtts(lefthikey, rel) > 0);+ Assert(BTreeTupleGetNAtts(lefthikey, rel) <=+ IndexRelationGetNumberOfKeyAttributes(rel));
if (PageAddItem(leftpage, (Item) lefthikey, itemsz, leftoff,
false, false) == InvalidOffsetNumber)
{
@@ -1458,7 +1500,6 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
xl_btree_split xlrec;
uint8 xlinfo;
XLogRecPtr recptr;
- bool loglhikey = false;
xlrec.level = ropaque->btpo.level;
xlrec.firstright = firstright;
@@ -1487,22 +1528,10 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
if (newitemonleft)
XLogRegisterBufData(0, (char *) newitem, MAXALIGN(newitemsz));
- /* Log left page */- if (!isleaf || indnatts != indnkeyatts)- {- /*- * We must also log the left page's high key. There are two- * reasons for that: right page's leftmost key is suppressed on- * non-leaf levels and in covering indexes included columns are- * truncated from high keys. Show it as belonging to the left- * page buffer, so that it is not stored if XLogInsert decides it- * needs a full-page image of the left page.- */- itemid = PageGetItemId(origpage, P_HIKEY);- item = (IndexTuple) PageGetItem(origpage, itemid);- XLogRegisterBufData(0, (char *) item, MAXALIGN(IndexTupleSize(item)));- loglhikey = true;- }+ /* Log left page. We must also log the left page's high key. */+ itemid = PageGetItemId(origpage, P_HIKEY);+ item = (IndexTuple) PageGetItem(origpage, itemid);+ XLogRegisterBufData(0, (char *) item, MAXALIGN(IndexTupleSize(item)));
/*
* Log the contents of the right page in the format understood by
@@ -1520,9 +1549,7 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
(char *) rightpage + ((PageHeader) rightpage)->pd_upper,
((PageHeader) rightpage)->pd_special - ((PageHeader) rightpage)->pd_upper);
- xlinfo = newitemonleft ?- (loglhikey ? XLOG_BTREE_SPLIT_L_HIGHKEY : XLOG_BTREE_SPLIT_L) :- (loglhikey ? XLOG_BTREE_SPLIT_R_HIGHKEY : XLOG_BTREE_SPLIT_R);+ xlinfo = newitemonleft ? XLOG_BTREE_SPLIT_L : XLOG_BTREE_SPLIT_R;
recptr = XLogInsert(RM_BTREE_ID, xlinfo);
PageSetLSN(origpage, recptr);
@@ -1575,10 +1602,31 @@ _bt_split(Relation rel, Buffer buf, Buffer cbuf, OffsetNumber firstright,
* righthand page, plus a boolean indicating whether the new tuple goes on
* the left or right page. The bool is necessary to disambiguate the case
* where firstright == newitemoff.
+ *+ * The top-level caller should specify disjunctpass as true. This makes+ * finding a split location refuse to consider candidate split points that+ * will result in a non-discriminating downlink for the right half of the+ * split in the parent page. If we fail to find a split point this way, we+ * call ourselves recursively without insisting on a disjunct split point.+ * This avoids cases where suffix truncation must add an extra heap TID to+ * the new pivot during a leaf page split. It also makes it far less likely+ * that any pivot tuples that have a heap attribute TID added despite our+ * best efforts will get used as a downlink a second time, in the+ * grandparent. There should only be downlinks with an extra heap TID+ * attribute in pages at higher levels of the tree when a very large+ * proportion of leaf tuples happen to be logical duplicates.+ *+ * The disjunctpass strategy may seem very aggressive, as it could lead to+ * each half of the split having very different amounts of free space.+ * Note, however, that logical duplicate downlinks in internal pages can+ * only help _bt_search() callers that pass a scantid argument (i.e.+ * UNIQUE_CHECK_NO inserters). Even there, the effect on performance ought+ * to be self-limiting.
*/
static OffsetNumber
_bt_findsplitloc(Relation rel,
Page page,
+ bool disjunctpass,
OffsetNumber newitemoff,
Size newitemsz,
bool *newitemonleft)
@@ -1661,6 +1709,14 @@ _bt_findsplitloc(Relation rel,
itemid = PageGetItemId(page, offnum);
itemsz = MAXALIGN(ItemIdGetLength(itemid)) + sizeof(ItemIdData);
+ /*+ * FIXME: This doesn't consider the case where the tuple to be+ * inserted is placed immediately before or after item.+ */+ if (disjunctpass && offnum > P_FIRSTDATAKEY(opaque) &&+ !_bt_split_isdisjunct(page, offnum))+ goto skipnondisjunct;+
/*
* Will the new item go to left or right of split?
*/
@@ -1688,6 +1744,8 @@ _bt_findsplitloc(Relation rel,
break;
}
+skipnondisjunct:+
olddataitemstoleft += itemsz;
}
@@ -1699,18 +1757,80 @@ _bt_findsplitloc(Relation rel,
if (newitemoff > maxoff && !goodenoughfound)
_bt_checksplitloc(&state, newitemoff, false, olddataitemstotal, 0);
- /*- * I believe it is not possible to fail to find a feasible split, but just- * in case ...- */
if (!state.have_split)
+ {+ /*+ * We couldn't find a split point using disjunctpass strategy. Take+ * a second pass at it, this time only weighing space utilization on+ * the page that we're about to split.+ */+ if (disjunctpass)+ return _bt_findsplitloc(rel, page, false, newitemoff, newitemsz,+ newitemonleft);+ /*+ * I believe it is not possible to fail to find a feasible split, but+ * just in case ...+ */
elog(ERROR, "could not find a feasible split point for index \"%s\"",
RelationGetRelationName(rel));
+ }
*newitemonleft = state.newitemonleft;
return state.firstright;
}
+/*+ * _bt_split_isdisjunct - does split point separate key space well?+ *+ * firstoldonright is a candidate split point.+ *+ * Checks if point between the would-be last left and first right tuples on the+ * hypothetical post-split halves are disjunct.+ */+static bool+_bt_split_isdisjunct(Page page, OffsetNumber firstoldonright)+{+ BTPageOpaque opaque;+ ItemId itemid;+ IndexTuple itup;++ opaque = (BTPageOpaque) PageGetSpecialPointer(page);+ itemid = PageGetItemId(page, firstoldonright);+ itup = (IndexTuple) PageGetItem(page, itemid);++ if (P_ISLEAF(opaque))+ {+ IndexTuple lastitup;+ char *curTup;+ char *lastTup;+ Size datasz;++ /*+ * Use memcmp() to cheaply approximate real comparisons.+ *+ * XXX: This clearly needs more work, especially to make things like+ * TPC-C's order_lines pkey make better use of suffix truncation.+ *+ * FIXME: What about covering index pivot tuples? Clearly using+ * memcmp() there does not approximate comparing key attributes very+ * well at all.+ */+ itemid = PageGetItemId(page, OffsetNumberPrev(firstoldonright));+ lastitup = (IndexTuple) PageGetItem(page, itemid);+ lastTup = (char *) lastitup + IndexInfoFindDataOffset(lastitup->t_info);+ curTup = (char *) itup + IndexInfoFindDataOffset(itup->t_info);+ datasz = IndexTupleSize(itup) - IndexInfoFindDataOffset(itup->t_info);++ if (IndexTupleSize(lastitup) == IndexTupleSize(itup) &&+ memcmp(lastTup, curTup, datasz) == 0)+ return false;+ }+ else if (BTreeTupleGetHeapTID(itup) != NULL)+ return false;++ return true;+}+
/*
* Subroutine to analyze a particular possible split choice (ie, firstright
* and newitemonleft settings), and record the best split so far in *state.
@@ -1761,6 +1881,9 @@ _bt_checksplitloc(FindSplitData *state,
*/
leftfree -= firstrightitemsz;
+ /* Charge for case where a heap TID is added */+ leftfree -= MAXALIGN(sizeof(ItemPointerData));+
/* account for the new item */
if (newitemonleft)
leftfree -= (int) state->newitemsz;
@@ -2210,7 +2333,8 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
/*
* insert the right page pointer into the new root page.
*/
- Assert(BTreeTupleGetNAtts(right_item, rel) ==+ Assert(BTreeTupleGetNAtts(right_item, rel) > 0);+ Assert(BTreeTupleGetNAtts(right_item, rel) <=
IndexRelationGetNumberOfKeyAttributes(rel));
if (PageAddItem(rootpage, (Item) right_item, right_item_sz, P_FIRSTKEY,
false, false) == InvalidOffsetNumber)
@@ -2322,8 +2446,8 @@ _bt_pgaddtup(Page page,
/*
* _bt_isequal - used in _bt_doinsert in check for duplicates.
*
- * This is very similar to _bt_compare, except for NULL handling.- * Rule is simple: NOT_NULL not equal NULL, NULL not equal NULL too.+ * This is very similar to _bt_compare, except for NULL and negative infinity+ * handling. Rule is simple: NOT_NULL not equal NULL, NULL not equal NULL too.
*/
static bool
_bt_isequal(TupleDesc itupdesc, Page page, OffsetNumber offnum,
@@ -2337,12 +2461,6 @@ _bt_isequal(TupleDesc itupdesc, Page page, OffsetNumber offnum,
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
- /*- * It's okay that we might perform a comparison against a truncated page- * high key when caller needs to determine if _bt_check_unique scan must- * continue on to the next page. Caller never asks us to compare non-key- * attributes within an INCLUDE index.- */
for (i = 1; i <= keysz; i++)
{
AttrNumber attno;
diff --git a/src/backend/access/nbtree/nbtpage.c b/src/backend/access/nbtree/nbtpage.cindex 4082103fe2..f63615341c 100644--- a/src/backend/access/nbtree/nbtpage.c+++ b/src/backend/access/nbtree/nbtpage.c@@ -1421,10 +1421,12 @@ _bt_pagedel(Relation rel, Buffer buf)
/* we need an insertion scan key for the search, so build one */
itup_scankey = _bt_mkscankey(rel, targetkey);
- /* find the leftmost leaf page containing this key */+ /* get stack to leaf page by searching index */
stack = _bt_search(rel,
- IndexRelationGetNumberOfKeyAttributes(rel),- itup_scankey, false, &lbuf, BT_READ, NULL);+ BTreeTupleGetNAtts(targetkey, rel),+ itup_scankey,+ BTreeTupleGetHeapTID(targetkey), false,+ &lbuf, BT_READ, NULL);
/* don't need a pin on the page */
_bt_relbuf(rel, lbuf);
diff --git a/src/backend/access/nbtree/nbtsearch.c b/src/backend/access/nbtree/nbtsearch.cindex 0bcfa10b86..b5e22bcb09 100644--- a/src/backend/access/nbtree/nbtsearch.c+++ b/src/backend/access/nbtree/nbtsearch.c@@ -69,11 +69,13 @@ _bt_drop_lock_and_maybe_pin(IndexScanDesc scan, BTScanPos sp)
/*
- * _bt_search() -- Search the tree for a particular scankey,+ * _bt_search() -- Search the tree for a particular scankey + scantid,
* or more precisely for the first leaf page it could be on.
*
* The passed scankey must be an insertion-type scankey (see nbtree/README),
- * but it can omit the rightmost column(s) of the index.+ * but it can omit the rightmost column(s) of the index. The scantid+ * argument may also be omitted (caller passes NULL), since it's logically+ * the "real" rightmost attribute.
*
* When nextkey is false (the usual case), we are looking for the first
* item >= scankey. When nextkey is true, we are looking for the first
@@ -94,8 +96,8 @@ _bt_drop_lock_and_maybe_pin(IndexScanDesc scan, BTScanPos sp)
* any incomplete splits encountered during the search will be finished.
*/
BTStack
-_bt_search(Relation rel, int keysz, ScanKey scankey, bool nextkey,- Buffer *bufP, int access, Snapshot snapshot)+_bt_search(Relation rel, int keysz, ScanKey scankey, ItemPointer scantid,+ bool nextkey, Buffer *bufP, int access, Snapshot snapshot)
{
BTStack stack_in = NULL;
@@ -130,7 +132,7 @@ _bt_search(Relation rel, int keysz, ScanKey scankey, bool nextkey,
* if the leaf page is split and we insert to the parent page). But
* this is a good opportunity to finish splits of internal pages too.
*/
- *bufP = _bt_moveright(rel, *bufP, keysz, scankey, nextkey,+ *bufP = _bt_moveright(rel, *bufP, keysz, scankey, scantid, nextkey,
(access == BT_WRITE), stack_in,
BT_READ, snapshot);
@@ -144,7 +146,8 @@ _bt_search(Relation rel, int keysz, ScanKey scankey, bool nextkey,
* Find the appropriate item on the internal page, and get the child
* page that it points to.
*/
- offnum = _bt_binsrch(rel, *bufP, keysz, scankey, nextkey);+ offnum = _bt_binsrch(rel, *bufP, keysz, scankey, scantid,+ P_FIRSTDATAKEY(opaque), nextkey);
itemid = PageGetItemId(page, offnum);
itup = (IndexTuple) PageGetItem(page, itemid);
blkno = BTreeInnerTupleGetDownLink(itup);
@@ -157,8 +160,8 @@ _bt_search(Relation rel, int keysz, ScanKey scankey, bool nextkey,
* downlink (block) to uniquely identify the index entry, in case it
* moves right while we're working lower in the tree. See the paper
* by Lehman and Yao for how this is detected and handled. (We use the
- * child link to disambiguate duplicate keys in the index -- Lehman- * and Yao disallow duplicate keys.)+ * child link to disambiguate duplicate keys in the index, which is+ * faster than comparing the keys themselves.)
*/
new_stack = (BTStack) palloc(sizeof(BTStackData));
new_stack->bts_blkno = par_blkno;
@@ -215,6 +218,7 @@ _bt_moveright(Relation rel,
Buffer buf,
int keysz,
ScanKey scankey,
+ ItemPointer scantid,
bool nextkey,
bool forupdate,
BTStack stack,
@@ -275,7 +279,7 @@ _bt_moveright(Relation rel,
continue;
}
- if (P_IGNORE(opaque) || _bt_compare(rel, keysz, scankey, page, P_HIKEY) >= cmpval)+ if (P_IGNORE(opaque) || _bt_compare(rel, keysz, scankey, scantid, page, P_HIKEY) >= cmpval)
{
/* step right one page */
buf = _bt_relandgetbuf(rel, buf, opaque->btpo_next, access);
@@ -307,6 +311,12 @@ _bt_moveright(Relation rel,
* particular, this means it is possible to return a value 1 greater than the
* number of keys on the page, if the scankey is > all keys on the page.)
*
+ * Caller passes own low value for binary search. This can be used to+ * resume a partial binary search without repeated effort. _bt_check_unique+ * callers use this to avoid repeated work. This only works when a buffer+ * lock is held throughout, and we're passed a leaf page both times, and+ * nextkey is false.+ *
* On an internal (non-leaf) page, _bt_binsrch() returns the OffsetNumber
* of the last key < given scankey, or last key <= given scankey if nextkey
* is true. (Since _bt_compare treats the first data key of such a page as
@@ -324,19 +334,19 @@ _bt_binsrch(Relation rel,
Buffer buf,
int keysz,
ScanKey scankey,
+ ItemPointer scantid,+ OffsetNumber low,
bool nextkey)
{
Page page;
BTPageOpaque opaque;
- OffsetNumber low,- high;+ OffsetNumber high;
int32 result,
cmpval;
page = BufferGetPage(buf);
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
- low = P_FIRSTDATAKEY(opaque);
high = PageGetMaxOffsetNumber(page);
/*
@@ -371,7 +381,7 @@ _bt_binsrch(Relation rel,
/* We have low <= mid < high, so mid points at a real slot */
- result = _bt_compare(rel, keysz, scankey, page, mid);+ result = _bt_compare(rel, keysz, scankey, scantid, page, mid);
if (result >= cmpval)
low = mid + 1;
@@ -401,20 +411,8 @@ _bt_binsrch(Relation rel,
/*----------
* _bt_compare() -- Compare scankey to a particular tuple on the page.
*
- * The passed scankey must be an insertion-type scankey (see nbtree/README),- * but it can omit the rightmost column(s) of the index.- *- * keysz: number of key conditions to be checked (might be less than the- * number of index columns!)- * page/offnum: location of btree item to be compared to.- *- * This routine returns:- * <0 if scankey < tuple at offnum;- * 0 if scankey == tuple at offnum;- * >0 if scankey > tuple at offnum.- * NULLs in the keys are treated as sortable values. Therefore- * "equality" does not necessarily mean that the item should be- * returned to the caller as a matching key!+ * Convenience wrapper for _bt_tuple_compare() callers that want to compare+ * an offset on a particular page.
*
* CRUCIAL NOTE: on a non-leaf page, the first data key is assumed to be
* "minus infinity": this routine will always claim it is less than the
@@ -428,24 +426,66 @@ int32
_bt_compare(Relation rel,
int keysz,
ScanKey scankey,
+ ItemPointer scantid,
Page page,
OffsetNumber offnum)
{
- TupleDesc itupdesc = RelationGetDescr(rel);
BTPageOpaque opaque = (BTPageOpaque) PageGetSpecialPointer(page);
IndexTuple itup;
- int i;
Assert(_bt_check_natts(rel, page, offnum));
/*
* Force result ">" if target item is first data item on an internal page
* --- see NOTE above.
+ *+ * A minus infinity key has all attributes truncated away, so this test is+ * redundant with the minus infinity attribute tie-breaker. However, the+ * number of attributes in minus infinity tuples was not explicitly+ * represented as 0 until PostgreSQL v11, so an explicit offnum test is+ * still required.
*/
if (!P_ISLEAF(opaque) && offnum == P_FIRSTDATAKEY(opaque))
return 1;
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
+ return _bt_tuple_compare(rel, keysz, scankey, scantid, itup);+}++/*----------+ * _bt_tuple_compare() -- Compare scankey to a particular tuple.+ *+ * The passed scankey must be an insertion-type scankey (see nbtree/README),+ * but it can omit the rightmost column(s) of the index.+ *+ * keysz: number of key conditions to be checked (might be less than the+ * number of index columns!)+ * page/offnum: location of btree item to be compared to.+ *+ * This routine returns:+ * <0 if scankey < tuple at offnum;+ * 0 if scankey == tuple at offnum;+ * >0 if scankey > tuple at offnum.+ * NULLs in the keys are treated as sortable values. Therefore+ * "equality" does not necessarily mean that the item should be+ * returned to the caller as a matching key!+ *----------+ */+int32+_bt_tuple_compare(Relation rel,+ int keysz,+ ScanKey scankey,+ ItemPointer scantid,+ IndexTuple itup)+{+ TupleDesc itupdesc = RelationGetDescr(rel);+ ItemPointer heapTid;+ int ntupatts;+ int ncmpkey;+ int i;++ Assert(keysz <= IndexRelationGetNumberOfKeyAttributes(rel));+ ntupatts = BTreeTupleGetNAtts(itup, rel);
/*
* The scan key is set up with the attribute number associated with each
@@ -459,7 +499,8 @@ _bt_compare(Relation rel,
* _bt_first).
*/
- for (i = 1; i <= keysz; i++)+ ncmpkey = Min(ntupatts, keysz);+ for (i = 1; i <= ncmpkey; i++)
{
Datum datum;
bool isNull;
@@ -510,8 +551,31 @@ _bt_compare(Relation rel,
scankey++;
}
- /* if we get here, the keys are equal */- return 0;+ /*+ * Use the number of attributes as a tie-breaker, in order to treat+ * truncated attributes in index as minus infinity.+ */+ if (keysz > ntupatts)+ return 1;++ /* If caller provided no heap TID tie-breaker for scan, they're equal */+ if (!scantid)+ return 0;++ /*+ * Although it isn't counted as an attribute by BTreeTupleGetNAtts(), heap+ * TID is an implicit final key attribute that ensures that all index+ * tuples have a distinct set of key attribute values.+ *+ * This is often truncated away in pivot tuples, which makes the attribute+ * value implicitly negative infinity.+ */+ heapTid = BTreeTupleGetHeapTID(itup);+ if (!heapTid)+ return 1;++ /* Deliberately invert the order, since TIDs "sort DESC" */+ return ItemPointerCompare(heapTid, scantid);
}
/*
@@ -540,6 +604,7 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
Relation rel = scan->indexRelation;
BTScanOpaque so = (BTScanOpaque) scan->opaque;
Buffer buf;
+ BTPageOpaque opaque;
BTStack stack;
OffsetNumber offnum;
StrategyNumber strat;
@@ -547,6 +612,8 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
bool goback;
ScanKey startKeys[INDEX_MAX_KEYS];
ScanKeyData scankeys[INDEX_MAX_KEYS];
+ ItemPointer scantid;+ ItemPointerData minscantid;
ScanKeyData notnullkeys[INDEX_MAX_KEYS];
int keysCount = 0;
int i;
@@ -796,6 +863,7 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
* scankeys[] array, using the keys identified by startKeys[].
*/
Assert(keysCount <= INDEX_MAX_KEYS);
+ scantid = NULL;
for (i = 0; i < keysCount; i++)
{
ScanKey cur = startKeys[i];
@@ -932,6 +1000,30 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
}
}
+ /*+ * When all key attributes will be in insertion scankey, manufacture+ * sentinel scan tid that's less than any possible heap TID in the index.+ * This is still greater than minus infinity to _bt_compare, allowing+ * _bt_search to follow a downlink with scankey-equal attributes, but a+ * truncated-away heap TID.+ *+ * If we didn't do this then affected index scans would have to+ * unnecessarily visit an extra page before moving right to the page they+ * should have landed on from the parent in the first place.+ *+ * (Note that implementing this by adding hard-coding to _bt_compare is+ * unworkable, since some _bt_search callers need to re-find a leaf page+ * using the page's high key.)+ */+ if (keysCount >= IndexRelationGetNumberOfKeyAttributes(rel))+ {+ scantid = &minscantid;++ /* Heap TID attribute uses DESC ordering */+ ItemPointerSetBlockNumber(scantid, InvalidBlockNumber);+ ItemPointerSetOffsetNumber(scantid, InvalidOffsetNumber);+ }+
/*----------
* Examine the selected initial-positioning strategy to determine exactly
* where we need to start the scan, and set flag variables to control the
@@ -1024,11 +1116,11 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
}
/*
- * Use the manufactured insertion scan key to descend the tree and- * position ourselves on the target leaf page.+ * Use the manufactured insertion scan key (and possibly a scantid) to+ * descend the tree and position ourselves on the target leaf page.
*/
- stack = _bt_search(rel, keysCount, scankeys, nextkey, &buf, BT_READ,- scan->xs_snapshot);+ stack = _bt_search(rel, keysCount, scankeys, scantid, nextkey, &buf,+ BT_READ, scan->xs_snapshot);
/* don't need to keep the stack around... */
_bt_freestack(stack);
@@ -1057,7 +1149,9 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
_bt_initialize_more_data(so, dir);
/* position to the precise item on the page */
- offnum = _bt_binsrch(rel, buf, keysCount, scankeys, nextkey);+ opaque = (BTPageOpaque) PageGetSpecialPointer(BufferGetPage(buf));+ offnum = _bt_binsrch(rel, buf, keysCount, scankeys, scantid,+ P_FIRSTDATAKEY(opaque), nextkey);
/*
* If nextkey = false, we are positioned at the first item >= scan key, or
diff --git a/src/backend/access/nbtree/nbtsort.c b/src/backend/access/nbtree/nbtsort.cindex 16f5755777..24bae8a454 100644--- a/src/backend/access/nbtree/nbtsort.c+++ b/src/backend/access/nbtree/nbtsort.c@@ -796,8 +796,6 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
OffsetNumber last_off;
Size pgspc;
Size itupsz;
- int indnatts = IndexRelationGetNumberOfAttributes(wstate->index);- int indnkeyatts = IndexRelationGetNumberOfKeyAttributes(wstate->index);
/*
* This is a handy place to check for cancel interrupts during the btree
@@ -880,19 +878,28 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
ItemIdSetUnused(ii); /* redundant */
((PageHeader) opage)->pd_lower -= sizeof(ItemIdData);
- if (indnkeyatts != indnatts && P_ISLEAF(opageop))+ if (P_ISLEAF(opageop))
{
- IndexTuple truncated;- Size truncsz;+ IndexTuple lastleft;+ IndexTuple truncated;+ Size truncsz;
/*
- * Truncate any non-key attributes from high key on leaf level- * (i.e. truncate on leaf level if we're building an INCLUDE- * index). This is only done at the leaf level because downlinks- * in internal pages are either negative infinity items, or get- * their contents from copying from one level down. See also:+ * Truncate away any unneeded attributes from high key on leaf+ * level. This is only done at the leaf level because downlinks in+ * internal pages are either negative infinity items, or get their+ * contents from copying from one level down. See also:
* _bt_split().
*
+ * We don't try to bias our choice of split point to make it more+ * likely that _bt_suffix_truncate() can truncate away more+ * attributes, whereas the split point passed to _bt_split() is+ * chosen much more delicately. Suffix truncation is mostly useful+ * because it can greatly improve space utilization for workloads+ * with random insertions. It doesn't seem worthwhile to add+ * complex logic for choosing a split point here for a benefit that+ * is bound to be much smaller.+ *
* Since the truncated tuple is probably smaller than the
* original, it cannot just be copied in place (besides, we want
* to actually save space on the leaf page). We delete the
@@ -905,7 +912,10 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
* the latter portion of the space occupied by the original tuple.
* This is fairly cheap.
*/
- truncated = _bt_nonkey_truncate(wstate->index, oitup);+ ii = PageGetItemId(opage, OffsetNumberPrev(last_off));+ lastleft = (IndexTuple) PageGetItem(opage, ii);++ truncated = _bt_suffix_truncate(wstate->index, lastleft, oitup);
truncsz = IndexTupleSize(truncated);
PageIndexTupleDelete(opage, P_HIKEY);
_bt_sortaddtup(opage, truncsz, truncated, P_HIKEY);
@@ -924,8 +934,9 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
if (state->btps_next == NULL)
state->btps_next = _bt_pagestate(wstate, state->btps_level + 1);
- Assert(BTreeTupleGetNAtts(state->btps_minkey, wstate->index) ==- IndexRelationGetNumberOfKeyAttributes(wstate->index) ||+ Assert((BTreeTupleGetNAtts(state->btps_minkey, wstate->index) <=+ IndexRelationGetNumberOfKeyAttributes(wstate->index) &&+ BTreeTupleGetNAtts(state->btps_minkey, wstate->index) > 0) ||
P_LEFTMOST(opageop));
Assert(BTreeTupleGetNAtts(state->btps_minkey, wstate->index) == 0 ||
!P_LEFTMOST(opageop));
@@ -970,7 +981,7 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
* the first item for a page is copied from the prior page in the code
* above. Since the minimum key for an entire level is only used as a
* minus infinity downlink, and never as a high key, there is no need to
- * truncate away non-key attributes at this point.+ * truncate away suffix attributes at this point.
*/
if (last_off == P_HIKEY)
{
@@ -1029,8 +1040,9 @@ _bt_uppershutdown(BTWriteState *wstate, BTPageState *state)
}
else
{
- Assert(BTreeTupleGetNAtts(s->btps_minkey, wstate->index) ==- IndexRelationGetNumberOfKeyAttributes(wstate->index) ||+ Assert((BTreeTupleGetNAtts(s->btps_minkey, wstate->index) <=+ IndexRelationGetNumberOfKeyAttributes(wstate->index) &&+ BTreeTupleGetNAtts(s->btps_minkey, wstate->index) > 0) ||
P_LEFTMOST(opaque));
Assert(BTreeTupleGetNAtts(s->btps_minkey, wstate->index) == 0 ||
!P_LEFTMOST(opaque));
@@ -1127,6 +1139,8 @@ _bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
}
else if (itup != NULL)
{
+ int32 compare = 0;+
for (i = 1; i <= keysz; i++)
{
SortSupport entry;
@@ -1134,7 +1148,6 @@ _bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
attrDatum2;
bool isNull1,
isNull2;
- int32 compare;
entry = sortKeys + i - 1;
attrDatum1 = index_getattr(itup, i, tupdes, &isNull1);
@@ -1151,6 +1164,22 @@ _bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
else if (compare < 0)
break;
}
++ /*+ * If key values are equal, we sort on ItemPointer. This is+ * required for btree indexes, since heap TID is treated as an+ * implicit last key attribute in order to ensure that all keys+ * in the index are physically unique.+ *+ * Deliberately invert the order, since TIDs "sort DESC".+ */+ if (compare == 0)+ {+ compare = ItemPointerCompare(&itup2->t_tid, &itup->t_tid);+ Assert(compare != 0);+ if (compare > 0)+ load1 = false;+ }
}
else
load1 = false;
diff --git a/src/backend/access/nbtree/nbtutils.c b/src/backend/access/nbtree/nbtutils.cindex 4528e87c83..4c46decb1d 100644--- a/src/backend/access/nbtree/nbtutils.c+++ b/src/backend/access/nbtree/nbtutils.c@@ -49,6 +49,10 @@ static void _bt_mark_scankey_required(ScanKey skey);
static bool _bt_check_rowcompare(ScanKey skey,
IndexTuple tuple, TupleDesc tupdesc,
ScanDirection dir, bool *continuescan);
+static int _bt_leave_natts(Relation rel, IndexTuple lastleft,+ IndexTuple firstright);+static void _bt_set_median_tid(ItemPointer lastleft, ItemPointer firstright,+ ItemPointer pivotheaptid);
/*
@@ -56,27 +60,34 @@ static bool _bt_check_rowcompare(ScanKey skey,
* Build an insertion scan key that contains comparison data from itup
* as well as comparator routines appropriate to the key datatypes.
*
- * The result is intended for use with _bt_compare().+ * The result is intended for use with _bt_compare(). If itup has+ * undergone suffix truncation of key attributes, caller had better+ * pass BTreeTupleGetNAtts(itup, rel) as keysz to routines like+ * _bt_search() and _bt_compare() when using returned scan key. This+ * allows truncated attributes to participate in comparisons (truncated+ * attributes have implicit negative infinity values). Note that+ * _bt_compare() never treats a scan key as containing negative+ * infinity attributes.
*/
ScanKey
_bt_mkscankey(Relation rel, IndexTuple itup)
{
ScanKey skey;
TupleDesc itupdesc;
+ int tupnatts;
int indnatts PG_USED_FOR_ASSERTS_ONLY;
int indnkeyatts;
int16 *indoption;
int i;
itupdesc = RelationGetDescr(rel);
+ tupnatts = BTreeTupleGetNAtts(itup, rel);
indnatts = IndexRelationGetNumberOfAttributes(rel);
indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
indoption = rel->rd_indoption;
- Assert(indnkeyatts > 0);- Assert(indnkeyatts <= indnatts);- Assert(BTreeTupleGetNAtts(itup, rel) == indnatts ||- BTreeTupleGetNAtts(itup, rel) == indnkeyatts);+ Assert(tupnatts > 0);+ Assert(tupnatts <= indnatts);
/*
* We'll execute search using scan key constructed on key columns. Non-key
@@ -96,7 +107,21 @@ _bt_mkscankey(Relation rel, IndexTuple itup)
* comparison can be needed.
*/
procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
- arg = index_getattr(itup, i + 1, itupdesc, &null);++ /*+ * Truncated key attributes may not be represented in index tuple+ * due to suffix truncation. Keys built from truncated attributes+ * are defensively represented as NULL values, though they should+ * still not be allowed to participate in comparisons (caller must+ * be sure to pass a sane keysz to _bt_compare()).+ */+ if (i < tupnatts)+ arg = index_getattr(itup, i + 1, itupdesc, &null);+ else+ {+ arg = (Datum) 0;+ null = true;+ }
flags = (null ? SK_ISNULL : 0) | (indoption[i] << SK_BT_INDOPTION_SHIFT);
ScanKeyEntryInitializeWithInfo(&skey[i],
flags,
@@ -2083,38 +2108,218 @@ btproperty(Oid index_oid, int attno,
}
/*
- * _bt_nonkey_truncate() -- create tuple without non-key suffix attributes.+ * _bt_suffix_truncate() -- create tuple without unneeded suffix attributes.
*
- * Returns truncated index tuple allocated in caller's memory context, with key- * attributes copied from caller's itup argument. Currently, suffix truncation- * is only performed to create pivot tuples in INCLUDE indexes, but some day it- * could be generalized to remove suffix attributes after the first- * distinguishing key attribute.+ * Returns truncated pivot index tuple allocated in caller's memory context,+ * with key attributes copied from caller's firstright argument. If rel is+ * an INCLUDE index, non-key attributes will definitely be truncated away,+ * since they're not part of the key space. More aggressive suffix+ * truncation can take place when it's clear that the returned tuple does not+ * need one or more suffix key attributes. This is possible when there are+ * attributes that follow an attribute in firstright that is not equal to the+ * corresponding attribute in lastleft (equal according to an insertion scan+ * key).
*
- * Truncated tuple is guaranteed to be no larger than the original, which is- * important for staying under the 1/3 of a page restriction on tuple size.+ * Sometimes this routine will return a new pivot tuple that's larger than+ * firstright, because a new heap TID attribute had to be added to+ * distinguish lastleft from firstright. This should only happen when the+ * caller is in the process of splitting a leaf page that has many logical+ * duplicates, where it's unavoidable.
*
- * Note that returned tuple's t_tid offset will hold the number of attributes- * present, so the original item pointer offset is not represented. Caller- * should only change truncated tuple's downlink.+ * Note that returned tuple's t_tid offset will hold the number of+ * attributes present, so the original item pointer offset is not+ * represented. Caller should only change truncated tuple's downlink. Note+ * also that truncated key attributes are treated as containing "minus+ * infinity" values by _bt_compare()/_bt_tuple_compare(). Returned tuple is+ * guaranteed to be no larger than the original plus some extra space for a+ * possible extra heap TID tie-breaker attribute, which is important for+ * staying under the 1/3 of a page restriction on tuple size.
*/
IndexTuple
-_bt_nonkey_truncate(Relation rel, IndexTuple itup)+_bt_suffix_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright)
{
- int nkeyattrs = IndexRelationGetNumberOfKeyAttributes(rel);- IndexTuple truncated;+ TupleDesc itupdesc = RelationGetDescr(rel);+ int16 natts = IndexRelationGetNumberOfAttributes(rel);+ int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);+ int leavenatts;+ IndexTuple pivot;+ ItemPointer pivotheaptid;+ Size newsize;
/*
- * We should only ever truncate leaf index tuples, which must have both- * key and non-key attributes. It's never okay to truncate a second time.+ * We should only ever truncate leaf index tuples, which must have non-key+ * attributes in the case of INCLUDE indexes. It's never okay to truncate+ * a second time.+ */+ Assert(BTreeTupleGetNAtts(lastleft, rel) == natts);+ Assert(BTreeTupleGetNAtts(firstright, rel) == natts);++ /* Determine how many attributes must be left behind */+ leavenatts = _bt_leave_natts(rel, lastleft, firstright);++ if (leavenatts <= natts)+ {+ IndexTuple tidpivot;++ /*+ * Truncate away non-key attributes and/or key attributes. Do a+ * straight copy in the case where the only attribute to be "truncated+ * away" is the implicit heap TID key attribute (i.e. the case where we+ * can at least avoid adding an explicit heap TID attribute to new+ * pivot). We should only call index_truncate_tuple() when user+ * attributes need to be truncated.+ */+ if (leavenatts < natts)+ pivot = index_truncate_tuple(itupdesc, firstright, leavenatts);+ else+ pivot = CopyIndexTuple(firstright);++ /*+ * If there is a distinguishing key attribute within leavenatts, there+ * is no need to add an explicit heap TID attribute to new pivot.+ */+ if (leavenatts <= nkeyatts)+ {+ BTreeTupleSetNAtts(pivot, leavenatts);+ return pivot;+ }++ /*+ * Only non-key attributes could be truncated away from an INCLUDE+ * index's pivot tuple. They are not considered part of the key space,+ * so it's still necessary to add a heap TID attribute to the new pivot+ * tuple. Create enlarged copy of our truncated right tuple copy, to+ * fit heap TID.+ */+ Assert(natts < nkeyatts);+ newsize = IndexTupleSize(pivot) + MAXALIGN(sizeof(ItemPointerData));+ tidpivot = palloc0(newsize);+ memcpy(tidpivot, pivot, IndexTupleSize(pivot));+ pfree(pivot);+ pivot = tidpivot;+ }+ else+ {+ /*+ * No truncation was possible, since attributes are all equal. It's+ * necessary to add a heap TID attribute to the new pivot tuple.+ */+ Assert(natts == nkeyatts);+ newsize = IndexTupleSize(firstright) + MAXALIGN(sizeof(ItemPointerData));+ pivot = palloc0(newsize);+ memcpy(pivot, firstright, IndexTupleSize(firstright));+ }++ /*+ * Create enlarged copy of first right tuple to fit heap TID. We must+ * use heap TID as a unique-ifier in new pivot tuple, since no user key+ * attribute distinguishes which values belong on each side of the split+ * point.+ */+ pivot->t_info &= ~INDEX_SIZE_MASK;+ pivot->t_info |= newsize;++ /*+ * Generate an artificial heap TID value for the new pivot tuple. This+ * will be the median of the left and right heap TIDs, or a close+ * approximation.+ *+ * Note that we deliberately pass the firstright heap TID as low and the+ * lastleft heap TID as high, since the implicit heap TID attribute has+ * DESC sort order.+ *+ * Lehman and Yao require that the downlink to the right page, which is+ * to be inserted into the parent page in the second phase of a page+ * split be a strict lower bound on all current and future items on the+ * right page (this will be copied from the new high key for the left+ * side of the split). New pivot's heap TID attribute may occasionally+ * be equal to the the lastleft heap TID, but it must never be equal to+ * firstright's heap TID.
*/
- Assert(BTreeTupleGetNAtts(itup, rel) ==- IndexRelationGetNumberOfAttributes(rel));+ pivotheaptid = (ItemPointer) ((char *) pivot + newsize -+ MAXALIGN(sizeof(ItemPointerData)));+ _bt_set_median_tid(&firstright->t_tid, &lastleft->t_tid, pivotheaptid);+ Assert(ItemPointerCompare(&lastleft->t_tid, pivotheaptid) >= 0);+ Assert(ItemPointerCompare(&firstright->t_tid, pivotheaptid) < 0);- truncated = index_truncate_tuple(RelationGetDescr(rel), itup, nkeyattrs);- BTreeTupleSetNAtts(truncated, nkeyattrs);+ /* Mark tuple as containing all key attributes, plus TID attribute */+ BTreeTupleSetNAtts(pivot, nkeyatts);+ BTreeTupleSetAltHeapTID(pivot);++ return pivot;+}++/*+ * _bt_leave_natts - how many key attributes to leave when truncating.+ *+ * This can return a number of attributes that is one greater than the+ * number of key attributes for the index relation. This indicates that the+ * caller must use a heap TID as a unique-ifier in new pivot tuple.+ */+static int+_bt_leave_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright)+{+ int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);+ int leavenatts;+ ScanKey skey;++ skey = _bt_mkscankey(rel, firstright);++ /*+ * Even test nkeyatts (no truncated user attributes) case, since caller+ * cares about whether or not it can avoid appending a heap TID as a+ * unique-ifier+ */+ leavenatts = 1;+ for(;;)+ {+ if (leavenatts > nkeyatts)+ break;+ if (_bt_tuple_compare(rel, leavenatts, skey, NULL, lastleft) > 0)+ break;+ leavenatts++;+ }++ /* Can't leak memory here */+ _bt_freeskey(skey);++ return leavenatts;+}++/*+ * _bt_set_median_tid - set's item pointer to median TID value.+ */+static void+_bt_set_median_tid(ItemPointer low, ItemPointer high,+ ItemPointer pivotheaptid)+{+ uint64 lowblock, highblock, medianblock;+ uint32 lowoffset, highoffset, medianoffset;++ Assert(ItemPointerCompare(low, high) < 0);++ lowblock = ItemPointerGetBlockNumber(low);+ highblock = ItemPointerGetBlockNumber(high);++ lowoffset = ItemPointerGetOffsetNumber(low);+ highoffset = ItemPointerGetOffsetNumber(high);++ medianblock = (lowblock + highblock) / 2;+ if (medianblock >= highblock)+ {+ medianblock = highblock;++ /* Cannot allow result to equal low */+ if (medianblock == highblock)+ medianoffset = highoffset;+ else+ medianoffset = (lowoffset + highoffset) / 2;+ }+ else+ medianoffset = Max(lowoffset, highoffset) + 1;- return truncated;+ ItemPointerSetBlockNumber(pivotheaptid, medianblock);+ ItemPointerSetOffsetNumber(pivotheaptid, medianoffset);
}
/*
@@ -2137,6 +2342,7 @@ _bt_check_natts(Relation rel, Page page, OffsetNumber offnum)
int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
BTPageOpaque opaque = (BTPageOpaque) PageGetSpecialPointer(page);
IndexTuple itup;
+ int tupnatts;
/*
* We cannot reliably test a deleted or half-deleted page, since they have
@@ -2156,6 +2362,7 @@ _bt_check_natts(Relation rel, Page page, OffsetNumber offnum)
"BT_N_KEYS_OFFSET_MASK can't fit INDEX_MAX_KEYS");
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
+ tupnatts = BTreeTupleGetNAtts(itup, rel);
if (P_ISLEAF(opaque))
{
@@ -2165,7 +2372,7 @@ _bt_check_natts(Relation rel, Page page, OffsetNumber offnum)
* Leaf tuples that are not the page high key (non-pivot tuples)
* should never be truncated
*/
- return BTreeTupleGetNAtts(itup, rel) == natts;+ return tupnatts == natts;
}
else
{
@@ -2176,7 +2383,7 @@ _bt_check_natts(Relation rel, Page page, OffsetNumber offnum)
Assert(!P_RIGHTMOST(opaque));
/* Page high key tuple contains only key attributes */
- return BTreeTupleGetNAtts(itup, rel) == nkeyatts;+ return tupnatts > 0 && tupnatts <= nkeyatts;
}
}
else /* !P_ISLEAF(opaque) */
@@ -2209,7 +2416,7 @@ _bt_check_natts(Relation rel, Page page, OffsetNumber offnum)
* Tuple contains only key attributes despite on is it page high
* key or not
*/
- return BTreeTupleGetNAtts(itup, rel) == nkeyatts;+ return tupnatts > 0 && tupnatts <= nkeyatts;
}
}
diff --git a/src/backend/access/nbtree/nbtxlog.c b/src/backend/access/nbtree/nbtxlog.cindex 67a94cb80a..7c061e96d2 100644--- a/src/backend/access/nbtree/nbtxlog.c+++ b/src/backend/access/nbtree/nbtxlog.c@@ -202,7 +202,7 @@ btree_xlog_insert(bool isleaf, bool ismeta, XLogReaderState *record)
}
static void
-btree_xlog_split(bool onleft, bool lhighkey, XLogReaderState *record)+btree_xlog_split(bool onleft, XLogReaderState *record)
{
XLogRecPtr lsn = record->EndRecPtr;
xl_btree_split *xlrec = (xl_btree_split *) XLogRecGetData(record);
@@ -213,8 +213,6 @@ btree_xlog_split(bool onleft, bool lhighkey, XLogReaderState *record)
BTPageOpaque ropaque;
char *datapos;
Size datalen;
- IndexTuple left_hikey = NULL;- Size left_hikeysz = 0;
BlockNumber leftsib;
BlockNumber rightsib;
BlockNumber rnext;
@@ -248,20 +246,6 @@ btree_xlog_split(bool onleft, bool lhighkey, XLogReaderState *record)
_bt_restore_page(rpage, datapos, datalen);
- /*- * When the high key isn't present is the wal record, then we assume it to- * be equal to the first key on the right page. It must be from the leaf- * level.- */- if (!lhighkey)- {- ItemId hiItemId = PageGetItemId(rpage, P_FIRSTDATAKEY(ropaque));-- Assert(isleaf);- left_hikey = (IndexTuple) PageGetItem(rpage, hiItemId);- left_hikeysz = ItemIdGetLength(hiItemId);- }-
PageSetLSN(rpage, lsn);
MarkBufferDirty(rbuf);
@@ -284,6 +268,8 @@ btree_xlog_split(bool onleft, bool lhighkey, XLogReaderState *record)
OffsetNumber off;
IndexTuple newitem = NULL;
Size newitemsz = 0;
+ IndexTuple left_hikey = NULL;+ Size left_hikeysz = 0;
Page newlpage;
OffsetNumber leftoff;
@@ -298,13 +284,10 @@ btree_xlog_split(bool onleft, bool lhighkey, XLogReaderState *record)
}
/* Extract left hikey and its size (assuming 16-bit alignment) */
- if (lhighkey)- {- left_hikey = (IndexTuple) datapos;- left_hikeysz = MAXALIGN(IndexTupleSize(left_hikey));- datapos += left_hikeysz;- datalen -= left_hikeysz;- }+ left_hikey = (IndexTuple) datapos;+ left_hikeysz = MAXALIGN(IndexTupleSize(left_hikey));+ datapos += left_hikeysz;+ datalen -= left_hikeysz;
Assert(datalen == 0);
@@ -1003,16 +986,10 @@ btree_redo(XLogReaderState *record)
btree_xlog_insert(false, true, record);
break;
case XLOG_BTREE_SPLIT_L:
- btree_xlog_split(true, false, record);- break;- case XLOG_BTREE_SPLIT_L_HIGHKEY:- btree_xlog_split(true, true, record);+ btree_xlog_split(true, record);
break;
case XLOG_BTREE_SPLIT_R:
- btree_xlog_split(false, false, record);- break;- case XLOG_BTREE_SPLIT_R_HIGHKEY:- btree_xlog_split(false, true, record);+ btree_xlog_split(false, record);
break;
case XLOG_BTREE_VACUUM:
btree_xlog_vacuum(record);
diff --git a/src/backend/access/rmgrdesc/nbtdesc.c b/src/backend/access/rmgrdesc/nbtdesc.cindex 5c4457179d..667c906b2e 100644--- a/src/backend/access/rmgrdesc/nbtdesc.c+++ b/src/backend/access/rmgrdesc/nbtdesc.c@@ -35,8 +35,6 @@ btree_desc(StringInfo buf, XLogReaderState *record)
}
case XLOG_BTREE_SPLIT_L:
case XLOG_BTREE_SPLIT_R:
- case XLOG_BTREE_SPLIT_L_HIGHKEY:- case XLOG_BTREE_SPLIT_R_HIGHKEY:
{
xl_btree_split *xlrec = (xl_btree_split *) rec;
@@ -130,12 +128,6 @@ btree_identify(uint8 info)
case XLOG_BTREE_SPLIT_R:
id = "SPLIT_R";
break;
- case XLOG_BTREE_SPLIT_L_HIGHKEY:- id = "SPLIT_L_HIGHKEY";- break;- case XLOG_BTREE_SPLIT_R_HIGHKEY:- id = "SPLIT_R_HIGHKEY";- break;
case XLOG_BTREE_VACUUM:
id = "VACUUM";
break;
diff --git a/src/backend/storage/page/bufpage.c b/src/backend/storage/page/bufpage.cindex dfbda5458f..ffeb0624fe 100644--- a/src/backend/storage/page/bufpage.c+++ b/src/backend/storage/page/bufpage.c@@ -854,10 +854,8 @@ PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems)
* PageIndexTupleDelete is the best way. Delete the items in reverse
* order so we don't have to think about adjusting item numbers for
* previous deletions.
- *- * TODO: tune the magic number here
*/
- if (nitems <= 2)+ if (nitems <= 7)
{
while (--nitems >= 0)
PageIndexTupleDelete(page, itemnos[nitems]);
diff --git a/src/backend/utils/sort/tuplesort.c b/src/backend/utils/sort/tuplesort.cindex 9fb33b9035..2a0b64ad47 100644--- a/src/backend/utils/sort/tuplesort.c+++ b/src/backend/utils/sort/tuplesort.c@@ -4057,23 +4057,26 @@ comparetup_index_btree(const SortTuple *a, const SortTuple *b,
}
/*
- * If key values are equal, we sort on ItemPointer. This does not affect- * validity of the finished index, but it may be useful to have index- * scans in physical order.+ * If key values are equal, we sort on ItemPointer. This is required+ * for btree indexes, since heap TID is treated as an implicit last+ * key attribute in order to ensure that all keys in the index are+ * physically unique.+ *+ * Deliberately invert the order, since TIDs "sort DESC".
*/
{
BlockNumber blk1 = ItemPointerGetBlockNumber(&tuple1->t_tid);
BlockNumber blk2 = ItemPointerGetBlockNumber(&tuple2->t_tid);
if (blk1 != blk2)
- return (blk1 < blk2) ? -1 : 1;+ return (blk1 < blk2) ? 1 : -1;
}
{
OffsetNumber pos1 = ItemPointerGetOffsetNumber(&tuple1->t_tid);
OffsetNumber pos2 = ItemPointerGetOffsetNumber(&tuple2->t_tid);
if (pos1 != pos2)
- return (pos1 < pos2) ? -1 : 1;+ return (pos1 < pos2) ? 1 : -1;
}
return 0;
diff --git a/src/include/access/nbtree.h b/src/include/access/nbtree.hindex 3b9461c3b4..69cafa2195 100644--- a/src/include/access/nbtree.h+++ b/src/include/access/nbtree.h@@ -114,16 +114,27 @@ typedef struct BTMetaPageData
#define BTREE_METAPAGE 0 /* first page is meta */
#define BTREE_MAGIC 0x053162 /* magic number of btree pages */
-#define BTREE_VERSION 3 /* current version number */-#define BTREE_MIN_VERSION 2 /* minimal supported version number */+/* FIXME: Support versions 2 and 3 for the benefit of pg_upgrade users */+#define BTREE_VERSION 4 /* current version number */+#define BTREE_MIN_VERSION 4 /* minimal supported version number */
/*
* Maximum size of a btree index entry, including its tuple header.
*
* We actually need to be able to fit three items on every page,
* so restrict any one item to 1/3 the per-page available space.
+ *+ * There are rare cases where _bt_suffix_truncate() will need to enlarge+ * a heap index tuple to make space for a tie-breaker heap TID+ * attribute, which we account for here.
*/
#define BTMaxItemSize(page) \
+ MAXALIGN_DOWN((PageGetPageSize(page) - \+ MAXALIGN(SizeOfPageHeaderData + \+ 3*sizeof(ItemIdData) + \+ 3*MAXALIGN(sizeof(ItemPointerData))) - \+ MAXALIGN(sizeof(BTPageOpaqueData))) / 3)+#define BTMaxItemSizeOld(page) \
MAXALIGN_DOWN((PageGetPageSize(page) - \
MAXALIGN(SizeOfPageHeaderData + 3*sizeof(ItemIdData)) - \
MAXALIGN(sizeof(BTPageOpaqueData))) / 3)
@@ -204,12 +215,11 @@ typedef struct BTMetaPageData
* real offset (downlinks only need to store a block number). The offset
* field only stores the number of attributes when the INDEX_ALT_TID_MASK
* bit is set (we never assume that pivot tuples must explicitly store the
- * number of attributes, and currently do not bother storing the number of- * attributes unless indnkeyatts actually differs from indnatts).- * INDEX_ALT_TID_MASK is only used for pivot tuples at present, though it's- * possible that it will be used within non-pivot tuples in the future. Do- * not assume that a tuple with INDEX_ALT_TID_MASK set must be a pivot- * tuple.+ * number of attributes). INDEX_ALT_TID_MASK is only used for pivot tuples+ * at present, though it's possible that it will be used within non-pivot+ * tuples in the future. Do not assume that a tuple with INDEX_ALT_TID_MASK+ * set must be a pivot tuple. A pivot tuple must have INDEX_ALT_TID_MASK set+ * as of BTREE_VERSION 4, however.
*
* The 12 least significant offset bits are used to represent the number of
* attributes in INDEX_ALT_TID_MASK tuples, leaving 4 bits that are reserved
@@ -219,6 +229,8 @@ typedef struct BTMetaPageData
#define INDEX_ALT_TID_MASK INDEX_AM_RESERVED_BIT
#define BT_RESERVED_OFFSET_MASK 0xF000
#define BT_N_KEYS_OFFSET_MASK 0x0FFF
+/* Reserved to indicate if heap TID is represented at end of tuple */+#define BT_HEAP_TID_ATTR 0x1000
/* Get/set downlink block number */
#define BTreeInnerTupleGetDownLink(itup) \
@@ -241,14 +253,15 @@ typedef struct BTMetaPageData
} while(0)
/*
- * Get/set number of attributes within B-tree index tuple. Asserts should be- * removed when BT_RESERVED_OFFSET_MASK bits will be used.+ * Get/set number of attributes within B-tree index tuple.+ *+ * Note that this does not include an implicit tie-breaker heap-TID+ * attribute, if any.
*/
#define BTreeTupleGetNAtts(itup, rel) \
( \
(itup)->t_info & INDEX_ALT_TID_MASK ? \
( \
- AssertMacro((ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_RESERVED_OFFSET_MASK) == 0), \
ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_N_KEYS_OFFSET_MASK \
) \
: \
@@ -257,10 +270,42 @@ typedef struct BTMetaPageData
#define BTreeTupleSetNAtts(itup, n) \
do { \
(itup)->t_info |= INDEX_ALT_TID_MASK; \
- Assert(((n) & BT_RESERVED_OFFSET_MASK) == 0); \
ItemPointerSetOffsetNumber(&(itup)->t_tid, (n) & BT_N_KEYS_OFFSET_MASK); \
} while(0)
+/*+ * Get tie-breaker heap TID attribute, if any. Macro works with both pivot+ * and non-pivot tuples.+ *+ * Assumes that any tuple without INDEX_ALT_TID_MASK set has a t_tid that+ * points to the heap, and that all pivot tuples have INDEX_ALT_TID_MASK set+ * (since all pivot tuples must as of BTREE_VERSION 4). When non-pivot+ * tuples use the INDEX_ALT_TID_MASK representation in the future, they'll+ * probably also contain a heap TID at the end of the tuple. We avoid+ * assuming that a tuple with INDEX_ALT_TID_MASK set is necessarily a pivot+ * tuple.+ */+#define BTreeTupleGetHeapTID(itup) \+ ( \+ (itup)->t_info & INDEX_ALT_TID_MASK && \+ (ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_HEAP_TID_ATTR) != 0 ? \+ ( \+ (ItemPointer) (((char *) (itup) + IndexTupleSize(itup)) - \+ MAXALIGN(sizeof(ItemPointerData))) \+ ) \+ : (itup)->t_info & INDEX_ALT_TID_MASK ? NULL : (ItemPointer) &((itup)->t_tid) \+ )+/*+ * Set the heap TID attribute for a tuple that uses the INDEX_ALT_TID_MASK+ * representation (currently limited to pivot tuples)+ */+#define BTreeTupleSetAltHeapTID(itup) \+ do { \+ Assert((itup)->t_info & INDEX_ALT_TID_MASK); \+ ItemPointerSetOffsetNumber(&(itup)->t_tid, \+ ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) | BT_HEAP_TID_ATTR); \+ } while(0)+
/*
* Operator strategy numbers for B-tree have been moved to access/stratnum.h,
* because many places need to use them in ScanKeyInit() calls.
@@ -564,15 +609,18 @@ extern int _bt_pagedel(Relation rel, Buffer buf);
* prototypes for functions in nbtsearch.c
*/
extern BTStack _bt_search(Relation rel,
- int keysz, ScanKey scankey, bool nextkey,+ int keysz, ScanKey scankey, ItemPointer scantid, bool nextkey,
Buffer *bufP, int access, Snapshot snapshot);
extern Buffer _bt_moveright(Relation rel, Buffer buf, int keysz,
- ScanKey scankey, bool nextkey, bool forupdate, BTStack stack,- int access, Snapshot snapshot);+ ScanKey scankey, ItemPointer scantid, bool nextkey,+ bool forupdate, BTStack stack, int access, Snapshot snapshot);
extern OffsetNumber _bt_binsrch(Relation rel, Buffer buf, int keysz,
- ScanKey scankey, bool nextkey);+ ScanKey scankey, ItemPointer scantid, OffsetNumber low,+ bool nextkey);
extern int32 _bt_compare(Relation rel, int keysz, ScanKey scankey,
- Page page, OffsetNumber offnum);+ ItemPointer scantid, Page page, OffsetNumber offnum);+extern int32 _bt_tuple_compare(Relation rel, int keysz, ScanKey scankey,+ ItemPointer scantid, IndexTuple itup);
extern bool _bt_first(IndexScanDesc scan, ScanDirection dir);
extern bool _bt_next(IndexScanDesc scan, ScanDirection dir);
extern Buffer _bt_get_endpoint(Relation rel, uint32 level, bool rightmost,
@@ -605,7 +653,8 @@ extern bytea *btoptions(Datum reloptions, bool validate);
extern bool btproperty(Oid index_oid, int attno,
IndexAMProperty prop, const char *propname,
bool *res, bool *isnull);
-extern IndexTuple _bt_nonkey_truncate(Relation rel, IndexTuple itup);+extern IndexTuple _bt_suffix_truncate(Relation rel, IndexTuple lastleft,+ IndexTuple firstright);
extern bool _bt_check_natts(Relation rel, Page page, OffsetNumber offnum);
/*
diff --git a/src/include/access/nbtxlog.h b/src/include/access/nbtxlog.hindex 819373031c..d0fd4229d2 100644--- a/src/include/access/nbtxlog.h+++ b/src/include/access/nbtxlog.h@@ -28,8 +28,8 @@
#define XLOG_BTREE_INSERT_META 0x20 /* same, plus update metapage */
#define XLOG_BTREE_SPLIT_L 0x30 /* add index tuple with split */
#define XLOG_BTREE_SPLIT_R 0x40 /* as above, new item on right */
-#define XLOG_BTREE_SPLIT_L_HIGHKEY 0x50 /* as above, include truncated highkey */-#define XLOG_BTREE_SPLIT_R_HIGHKEY 0x60 /* as above, include truncated highkey */+/* 0x50: unused */+/* 0x60: unused */
#define XLOG_BTREE_DELETE 0x70 /* delete leaf index tuples for a page */
#define XLOG_BTREE_UNLINK_PAGE 0x80 /* delete a half-dead page */
#define XLOG_BTREE_UNLINK_PAGE_META 0x90 /* same, and update metapage */
@@ -82,20 +82,16 @@ typedef struct xl_btree_insert
*
* Note: the four XLOG_BTREE_SPLIT xl_info codes all use this data record.
* The _L and _R variants indicate whether the inserted tuple went into the
- * left or right split page (and thus, whether newitemoff and the new item- * are stored or not). The _HIGHKEY variants indicate that we've logged- * explicitly left page high key value, otherwise redo should use right page- * leftmost key as a left page high key. _HIGHKEY is specified for internal- * pages where right page leftmost key is suppressed, and for leaf pages- * of covering indexes where high key have non-key attributes truncated.+ * left or right split page (and thus, whether newitemoff and the new item are+ * stored or not). We always explicitly log the left page high key value.
*
* Backup Blk 0: original page / new left page
*
* The left page's data portion contains the new item, if it's the _L variant.
- * (In the _R variants, the new item is one of the right page's tuples.)- * If level > 0, an IndexTuple representing the HIKEY of the left page- * follows. We don't need this on leaf pages, because it's the same as the- * leftmost key in the new right page.+ * In the _R variants, the new item is one of the right page's tuples. An+ * IndexTuple representing the HIKEY of the left page follows. We don't need+ * this on leaf pages, because it's the same as the leftmost key in the new+ * right page.
*
* Backup Blk 1: new right page
*
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.outindex dc6262be43..2c20cea4b9 100644--- a/src/test/regress/expected/join.out+++ b/src/test/regress/expected/join.out@@ -5896,8 +5896,8 @@ inner join j1 j2 on j1.id1 = j2.id1 and j1.id2 = j2.id2
where j1.id1 % 1000 = 1 and j2.id1 % 1000 = 1;
id1 | id2 | id1 | id2
-----+-----+-----+-----
- 1 | 1 | 1 | 1
1 | 2 | 1 | 2
+ 1 | 1 | 1 | 1
(2 rows)
reset enable_nestloop;
--
2.17.1
0004-Retail-IndexTuple-Deletion-with-TID-sorting-in-Leaf.patch (text/x-patch, 6.7K) download
From 223ba9983e713de76384c06e25f927000e85541a Mon Sep 17 00:00:00 2001
From: "Andrey V. Lepikhov" <a.lepikhov@postgrespro.ru>
Date: Wed, 15 Aug 2018 13:35:48 +0500
Subject: [PATCH 4/5] Retail-IndexTuple-Deletion-with-TID-sorting-in-Leaf
---
src/backend/access/nbtree/nbtree.c | 75 ++++++++++++++++++++----------
src/backend/commands/vacuumlazy.c | 8 ++--
src/include/access/genam.h | 2 +-
3 files changed, 55 insertions(+), 30 deletions(-)
diff --git a/src/backend/access/nbtree/nbtree.c b/src/backend/access/nbtree/nbtree.cindex 44f4d0dc9f..50378e0f51 100644--- a/src/backend/access/nbtree/nbtree.c+++ b/src/backend/access/nbtree/nbtree.c@@ -889,15 +889,19 @@ btbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
return stats;
}
-static int-tid_list_search(ItemPointer tid, ItemPointer tid_list, int ntid)-{- for (int i = 0; i < ntid; i++)- if (ItemPointerEquals(tid, &(tid_list[i])))- return i;- return -1;-}-+/*+ * Deletion of index entries pointing to heap tuples.+ *+ * Constraints:+ * 1. TID list info->dead_tuples arranged in ASC order.+ * 2. Logical duplicates of index tuples stored in DESC order.+ *+ * The function generates an insertion scan key and descent by btree for first+ * index tuple what satisfies scan key and last TID in info->dead_tuples list.+ * For the scan results it deletes all index entries, matched to the TID list.+ *+ * Result: a palloc'd struct containing statistical info.+ */
IndexTargetDeleteResult*
bttargetdelete(IndexTargetDeleteInfo *info,
IndexTargetDeleteResult *stats,
@@ -916,33 +920,33 @@ bttargetdelete(IndexTargetDeleteInfo *info,
int ndeletable = 0;
OffsetNumber deletable[MaxOffsetNumber];
IndexTuple itup;
+ int pos = info->last_dead_tuple;
if (stats == NULL)
stats = (IndexTargetDeleteResult *) palloc0(sizeof(IndexTargetDeleteResult));
+ /* Assemble scankey */
itup = index_form_tuple(RelationGetDescr(irel), values, isnull);
skey = _bt_mkscankey(irel, itup);
/* Descend the tree and position ourselves on the target leaf page. */
- stack = _bt_search(irel, keysCount, skey, false, &buf, BT_READ, NULL);- _bt_freestack(stack);+ stack = _bt_search(irel, keysCount, skey, &info->dead_tuples[pos], false, &buf, BT_READ, NULL);
/* To prepare tuple entries search across index pages */
Assert(BufferIsValid(buf));
- offnum = _bt_binsrch(irel, buf, keysCount, skey, false);
page = BufferGetPage(buf);
_bt_checkpage(irel, buf);
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
+ offnum = _bt_binsrch(irel, buf, keysCount, skey, &info->dead_tuples[pos], P_FIRSTDATAKEY(opaque), false);
for (;;)
{
int32 cmpval;
ItemId itemid;
IndexTuple itup;
- int pos;
/* Switch to the next page */
- if (P_IGNORE(opaque) || (offnum > PageGetMaxOffsetNumber(page)))+ if (offnum > PageGetMaxOffsetNumber(page))
{
/*
* Before unlocking index page we need to delete
@@ -967,20 +971,27 @@ bttargetdelete(IndexTargetDeleteInfo *info,
break;
/*
- * Switch to the next index page+ * Traverse to a next reliable index page
*/
- buf = _bt_relandgetbuf(irel, buf, opaque->btpo_next, BT_READ);+ buf = _bt_moveright(irel, buf, keysCount, skey, &info->dead_tuples[pos],+ false, true, stack, BT_READ, NULL);
page = BufferGetPage(buf);
_bt_checkpage(irel, buf);
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
- offnum = P_FIRSTDATAKEY(opaque);- continue;+ Assert(!P_IGNORE(opaque));+ /* Set offnum to first potentially interesting item */+ offnum = _bt_binsrch(irel, buf, keysCount, skey, &info->dead_tuples[pos], P_FIRSTDATAKEY(opaque), false);++ if (offnum > PageGetMaxOffsetNumber(page))+ break;+ else+ continue;
}
/*
* This index entry satisfied to the scan key?
*/
- cmpval = _bt_compare(irel, keysCount, skey, page, offnum);+ cmpval = _bt_compare(irel, keysCount, skey, NULL, page, offnum);
if (cmpval != 0)
/* End of index entries, satisfied to the scan key */
@@ -992,15 +1003,28 @@ bttargetdelete(IndexTargetDeleteInfo *info,
*/
itemid = PageGetItemId(page, offnum);
itup = (IndexTuple) PageGetItem(page, itemid);
- pos = tid_list_search(&(itup->t_tid), info->dead_tuples, info->num_dead_tuples);- if ((pos >= 0) && (!info->found_dead_tuples[pos]))+ /*+ * Search for next TID from presorted btree result comparable+ * to TID from presorted dead_tuples tid list+ */+ while (pos >= 0)
{
- /* index entry for TID of dead tuple is found */- deletable[ndeletable++] = offnum;- info->found_dead_tuples[pos] = true;+ int res = ItemPointerCompare(&(itup->t_tid), &info->dead_tuples[pos]);+ if ((res == 0) && (!info->found_dead_tuples[pos]))+ {+ /* index entry for TID of dead tuple is found */+ deletable[ndeletable++] = offnum;+ info->found_dead_tuples[pos] = true;+ }+ else if (res > 0)+ break;+ pos--;
}
+ if (pos < 0)+ break;+
offnum = OffsetNumberNext(offnum);
}
@@ -1017,7 +1041,8 @@ bttargetdelete(IndexTargetDeleteInfo *info,
stats->tuples_removed += ndeletable;
}
- /* Release scan key, unpin and unlock buffer */+ /* Release stack, scan key, unpin and unlock buffer */+ _bt_freestack(stack);
_bt_freeskey(skey);
_bt_relbuf(irel, buf);
diff --git a/src/backend/commands/vacuumlazy.c b/src/backend/commands/vacuumlazy.cindex 9a625b8dba..1b952fedd9 100644--- a/src/backend/commands/vacuumlazy.c+++ b/src/backend/commands/vacuumlazy.c@@ -1770,7 +1770,7 @@ quick_vacuum_index(Relation irel, Relation hrel,
econtext->ecxt_scantuple = slot;
/* Get tuple from heap */
- for (tnum = 0; tnum < vacrelstats->num_dead_tuples; tnum++)+ for (tnum = vacrelstats->num_dead_tuples-1; tnum >= 0; tnum--)
{
HeapTuple tuple;
Datum values[INDEX_MAX_KEYS];
@@ -1814,9 +1814,9 @@ quick_vacuum_index(Relation irel, Relation hrel,
* Make attempt to delete some index entries by one tree descent.
* We use only a part of TID list, which contains not found TID's.
*/
- ivinfo.dead_tuples = &(vacrelstats->dead_tuples[tnum]);- ivinfo.num_dead_tuples = vacrelstats->num_dead_tuples - tnum;- ivinfo.found_dead_tuples = found + tnum;+ ivinfo.dead_tuples = vacrelstats->dead_tuples;+ ivinfo.last_dead_tuple = tnum;+ ivinfo.found_dead_tuples = found;
index_target_delete(&ivinfo, &stats, values, isnull);
}
diff --git a/src/include/access/genam.h b/src/include/access/genam.hindex 79b386eebd..297d357a7d 100644--- a/src/include/access/genam.h+++ b/src/include/access/genam.h@@ -55,7 +55,7 @@ typedef struct IndexTargetDeleteInfo
{
Relation heapRelation;
Relation indexRelation; /* the index being vacuumed */
- int num_dead_tuples;+ int last_dead_tuple;
ItemPointer dead_tuples;
bool* found_dead_tuples;
} IndexTargetDeleteInfo;
--
2.17.1
0005-Background-Worker.patch (text/x-patch, 113.1K) download
From 3b21afe82bac9098249fb10bf3c3144fb01d871b Mon Sep 17 00:00:00 2001
From: "Andrey V. Lepikhov" <a.lepikhov@postgrespro.ru>
Date: Wed, 15 Aug 2018 13:45:22 +0500
Subject: [PATCH 5/5] Background Worker
---
.../postgres_fdw/expected/postgres_fdw.out | 30 +-
contrib/postgres_fdw/sql/postgres_fdw.sql | 4 +-
src/backend/access/heap/pruneheap.c | 5 +-
src/backend/access/nbtree/nbtinsert.c | 19 +
src/backend/access/nbtree/nbtree.c | 23 +-
src/backend/access/transam/xact.c | 4 +
src/backend/catalog/system_views.sql | 10 +
src/backend/commands/vacuumlazy.c | 46 +-
src/backend/postmaster/Makefile | 2 +-
src/backend/postmaster/bgheap.c | 1944 +++++++++++++++++
src/backend/postmaster/pgstat.c | 36 +
src/backend/postmaster/postmaster.c | 160 +-
src/backend/storage/buffer/bufmgr.c | 60 +-
src/backend/storage/buffer/localbuf.c | 24 +
src/backend/storage/ipc/ipci.c | 3 +
src/backend/storage/lmgr/lwlocknames.txt | 1 +
src/backend/storage/lmgr/proc.c | 5 +-
src/backend/tcop/postgres.c | 12 +
src/backend/utils/adt/pgstatfuncs.c | 2 +
src/backend/utils/hash/Makefile | 2 +-
src/backend/utils/hash/shash.c | 265 +++
src/backend/utils/init/miscinit.c | 3 +-
src/backend/utils/init/postinit.c | 11 +-
src/include/commands/progress.h | 8 +
src/include/commands/vacuum.h | 3 +
src/include/pgstat.h | 14 +-
src/include/postmaster/bgheap.h | 36 +
src/include/storage/buf_internals.h | 1 +
src/include/storage/bufmgr.h | 5 +-
src/include/storage/pmsignal.h | 2 +
src/include/utils/shash.h | 54 +
src/test/regress/expected/rules.out | 14 +-
src/test/regress/expected/triggers.out | 2 +-
src/test/regress/input/constraints.source | 12 +-
src/test/regress/output/constraints.source | 34 +-
src/test/regress/sql/rules.sql | 2 +-
src/test/regress/sql/triggers.sql | 2 +-
37 files changed, 2778 insertions(+), 82 deletions(-)
create mode 100644 src/backend/postmaster/bgheap.c
create mode 100644 src/backend/utils/hash/shash.c
create mode 100644 src/include/postmaster/bgheap.h
create mode 100644 src/include/utils/shash.h
diff --git a/contrib/postgres_fdw/expected/postgres_fdw.out b/contrib/postgres_fdw/expected/postgres_fdw.outindex f5498c62bd..622bea2a7d 100644--- a/contrib/postgres_fdw/expected/postgres_fdw.out+++ b/contrib/postgres_fdw/expected/postgres_fdw.out@@ -5496,19 +5496,25 @@ ALTER SERVER loopback OPTIONS (DROP extensions);
INSERT INTO ft2 (c1,c2,c3)
SELECT id, id % 10, to_char(id, 'FM00000') FROM generate_series(2001, 2010) id;
EXPLAIN (verbose, costs off)
-UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING *; -- can't be pushed down- QUERY PLAN ------------------------------------------------------------------------------------------------------------ Update on public.ft2- Output: c1, c2, c3, c4, c5, c6, c7, c8- Remote SQL: UPDATE "S 1"."T 1" SET c3 = $2 WHERE ctid = $1 RETURNING "C 1", c2, c3, c4, c5, c6, c7, c8- -> Foreign Scan on public.ft2- Output: c1, c2, NULL::integer, 'bar'::text, c4, c5, c6, c7, c8, ctid- Filter: (postgres_fdw_abs(ft2.c1) > 2000)- Remote SQL: SELECT "C 1", c2, c4, c5, c6, c7, c8, ctid FROM "S 1"."T 1" FOR UPDATE-(7 rows)+with updated AS ( UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING * ) SELECT * FROM updated ORDER BY c1; -- can't be pushed down+ QUERY PLAN +----------------------------------------------------------------------------------------------------------------------+ Sort+ Output: updated.c1, updated.c2, updated.c3, updated.c4, updated.c5, updated.c6, updated.c7, updated.c8+ Sort Key: updated.c1+ CTE updated+ -> Update on public.ft2+ Output: ft2.c1, ft2.c2, ft2.c3, ft2.c4, ft2.c5, ft2.c6, ft2.c7, ft2.c8+ Remote SQL: UPDATE "S 1"."T 1" SET c3 = $2 WHERE ctid = $1 RETURNING "C 1", c2, c3, c4, c5, c6, c7, c8+ -> Foreign Scan on public.ft2+ Output: ft2.c1, ft2.c2, NULL::integer, 'bar'::text, ft2.c4, ft2.c5, ft2.c6, ft2.c7, ft2.c8, ft2.ctid+ Filter: (postgres_fdw_abs(ft2.c1) > 2000)+ Remote SQL: SELECT "C 1", c2, c4, c5, c6, c7, c8, ctid FROM "S 1"."T 1" FOR UPDATE+ -> CTE Scan on updated+ Output: updated.c1, updated.c2, updated.c3, updated.c4, updated.c5, updated.c6, updated.c7, updated.c8+(13 rows)-UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING *;+with updated AS ( UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING * ) SELECT * FROM updated ORDER BY c1;
c1 | c2 | c3 | c4 | c5 | c6 | c7 | c8
------+----+-----+----+----+----+------------+----
2001 | 1 | bar | | | | ft2 |
diff --git a/contrib/postgres_fdw/sql/postgres_fdw.sql b/contrib/postgres_fdw/sql/postgres_fdw.sqlindex e1b955f3f0..6034e83108 100644--- a/contrib/postgres_fdw/sql/postgres_fdw.sql+++ b/contrib/postgres_fdw/sql/postgres_fdw.sql@@ -1137,8 +1137,8 @@ ALTER SERVER loopback OPTIONS (DROP extensions);
INSERT INTO ft2 (c1,c2,c3)
SELECT id, id % 10, to_char(id, 'FM00000') FROM generate_series(2001, 2010) id;
EXPLAIN (verbose, costs off)
-UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING *; -- can't be pushed down-UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING *;+with updated AS ( UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING * ) SELECT * FROM updated ORDER BY c1; -- can't be pushed down+with updated AS ( UPDATE ft2 SET c3 = 'bar' WHERE postgres_fdw_abs(c1) > 2000 RETURNING * ) SELECT * FROM updated ORDER BY c1;
EXPLAIN (verbose, costs off)
UPDATE ft2 SET c3 = 'baz'
FROM ft4 INNER JOIN ft5 ON (ft4.c1 = ft5.c1)
diff --git a/src/backend/access/heap/pruneheap.c b/src/backend/access/heap/pruneheap.cindex ebbafe3275..421dd607bb 100644--- a/src/backend/access/heap/pruneheap.c+++ b/src/backend/access/heap/pruneheap.c@@ -22,6 +22,7 @@
#include "catalog/catalog.h"
#include "miscadmin.h"
#include "pgstat.h"
+#include "postmaster/bgheap.h"
#include "storage/bufmgr.h"
#include "utils/snapmgr.h"
#include "utils/rel.h"
@@ -156,6 +157,7 @@ heap_page_prune_opt(Relation relation, Buffer buffer)
/* OK to prune */
(void) heap_page_prune(relation, buffer, OldestXmin, true, &ignore);
+ HeapCleanerSend(relation, BufferGetBlockNumber(buffer));
}
/* And release buffer lock */
@@ -411,7 +413,6 @@ heap_prune_chain(Relation relation, Buffer buffer, OffsetNumber rootoffnum,
heap_prune_record_dead(prstate, rootoffnum);
HeapTupleHeaderAdvanceLatestRemovedXid(htup,
&prstate->latestRemovedXid);
- ndeleted++;
}
/* Nothing more to do */
@@ -583,9 +584,9 @@ heap_prune_chain(Relation relation, Buffer buffer, OffsetNumber rootoffnum,
*/
for (i = 1; (i < nchain) && (chainitems[i - 1] != latestdead); i++)
{
- ndeleted++;
if (chainitems[i] == latestdead)
continue;
+ ndeleted++;
heap_prune_record_unused(prstate, chainitems[i]);
}
diff --git a/src/backend/access/nbtree/nbtinsert.c b/src/backend/access/nbtree/nbtinsert.cindex 7d0556b91d..c4cd941610 100644--- a/src/backend/access/nbtree/nbtinsert.c+++ b/src/backend/access/nbtree/nbtinsert.c@@ -493,6 +493,11 @@ _bt_check_unique(Relation rel, IndexTuple itup, Relation heapRel,
if (heap_hot_search(&htid, heapRel, SnapshotSelf, NULL))
{
/* Normal case --- it's still live */
+ elog(LOG, "It's still alive. off=%d BLK:(%d, %d)",+ offset,+ ItemPointerGetBlockNumber(&(itup->t_tid)),+ ItemPointerGetOffsetNumber(&(itup->t_tid))+ );
}
else
{
@@ -2522,7 +2527,21 @@ _bt_vacuum_one_page(Relation rel, Buffer buffer, Relation heapRel)
ItemId itemId = PageGetItemId(page, offnum);
if (ItemIdIsDead(itemId))
+ {
deletable[ndeletable++] = offnum;
+/* elog(LOG, "BACKEND remove (%d, %d) from index relation.",+ BufferGetBlockNumber(buffer),+ offnum+ );+ if (ItemIdHasStorage(itemId))+ {+ IndexTuple itup1 = (IndexTuple) PageGetItem(page, itemId);+ elog(LOG, "-> HEAP tid: (%d, %d)",+ ItemPointerGetBlockNumber(&(itup1->t_tid)),+ ItemPointerGetOffsetNumber(&(itup1->t_tid))+ );+ }*/+ }
}
if (ndeletable > 0)
diff --git a/src/backend/access/nbtree/nbtree.c b/src/backend/access/nbtree/nbtree.cindex 50378e0f51..a460cdfba6 100644--- a/src/backend/access/nbtree/nbtree.c+++ b/src/backend/access/nbtree/nbtree.c@@ -954,6 +954,7 @@ bttargetdelete(IndexTargetDeleteInfo *info,
*/
if (ndeletable > 0)
{
+ /* trade in our read lock for a write lock */
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
LockBufferForCleanup(buf);
@@ -1018,7 +1019,27 @@ bttargetdelete(IndexTargetDeleteInfo *info,
info->found_dead_tuples[pos] = true;
}
else if (res > 0)
+ {+ if (pos == info->last_dead_tuple)+ elog(LOG, "ByBreak");
break;
+ }/* else+ if (pos == info->last_dead_tuple)+ {+ OffsetNumber off1 = offnum-1;+ ItemId itemid1 = PageGetItemId(page, off1);+ IndexTuple itup1 = (IndexTuple) PageGetItem(page, itemid1);++ elog(LOG, "NEQ: off=%d/%d here: (%d, %d) search for: (%d, %d) pre: (%d, %d)", offnum,+ PageGetMaxOffsetNumber(page),+ ItemPointerGetBlockNumber(&(itup->t_tid)),+ ItemPointerGetOffsetNumber(&(itup->t_tid)),+ ItemPointerGetBlockNumber(&info->dead_tuples[pos]),+ ItemPointerGetOffsetNumber(&info->dead_tuples[pos]),+ ItemPointerGetBlockNumber(&(itup1->t_tid)),+ ItemPointerGetOffsetNumber(&(itup1->t_tid))+ );+ }*/
pos--;
}
@@ -1033,11 +1054,11 @@ bttargetdelete(IndexTargetDeleteInfo *info,
*/
if (ndeletable > 0)
{
+ /* trade in our read lock for a write lock */
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
LockBufferForCleanup(buf);
_bt_delitems_delete(irel, buf, deletable, ndeletable, hrel);
-
stats->tuples_removed += ndeletable;
}
diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.cindex 9aa63c8792..f5e70ae16d 100644--- a/src/backend/access/transam/xact.c+++ b/src/backend/access/transam/xact.c@@ -40,6 +40,7 @@
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
+#include "postmaster/bgheap.h"
#include "replication/logical.h"
#include "replication/logicallauncher.h"
#include "replication/origin.h"
@@ -2126,6 +2127,7 @@ CommitTransaction(void)
AtEOXact_Files(true);
AtEOXact_ComboCid();
AtEOXact_HashTables(true);
+ AtEOXact_BGHeap_tables(true);
AtEOXact_PgStat(true);
AtEOXact_Snapshot(true, false);
AtEOXact_ApplyLauncher(true);
@@ -2404,6 +2406,7 @@ PrepareTransaction(void)
AtEOXact_Files(true);
AtEOXact_ComboCid();
AtEOXact_HashTables(true);
+ AtEOXact_BGHeap_tables(true);
/* don't call AtEOXact_PgStat here; we fixed pgstat state above */
AtEOXact_Snapshot(true, true);
pgstat_report_xact_timestamp(0);
@@ -2606,6 +2609,7 @@ AbortTransaction(void)
AtEOXact_Files(false);
AtEOXact_ComboCid();
AtEOXact_HashTables(false);
+ AtEOXact_BGHeap_tables(true);
AtEOXact_PgStat(false);
AtEOXact_ApplyLauncher(false);
pgstat_report_xact_timestamp(0);
diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sqlindex 7251552419..17623b0f8d 100644--- a/src/backend/catalog/system_views.sql+++ b/src/backend/catalog/system_views.sql@@ -903,6 +903,16 @@ CREATE VIEW pg_stat_progress_vacuum AS
FROM pg_stat_get_progress_info('VACUUM') AS S
LEFT JOIN pg_database D ON S.datid = D.oid;
+CREATE VIEW pg_stat_progress_cleaner AS+ SELECT+ S.pid AS pid, S.datid AS datid, D.datname AS datname,+ S.relid AS relid,+ S.param1 AS relations, S.param2 AS queue_len, S.param3 AS false_hits,+ S.param4 AS cleaned_tuples, S.param5 AS vainly_cleaned_tuples,+ S.param6 AS missed_tuples+ FROM pg_stat_get_progress_info('CLEANER') AS S+ LEFT JOIN pg_database D ON S.datid = D.oid;+
CREATE VIEW pg_user_mappings AS
SELECT
U.oid AS umid,
diff --git a/src/backend/commands/vacuumlazy.c b/src/backend/commands/vacuumlazy.cindex 1b952fedd9..8620c07a2b 100644--- a/src/backend/commands/vacuumlazy.c+++ b/src/backend/commands/vacuumlazy.c@@ -158,8 +158,6 @@ static void lazy_scan_heap(Relation onerel, int options,
bool aggressive);
static void lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats);
static bool lazy_check_needs_freeze(Buffer buf, bool *hastup);
-static void quick_vacuum_index(Relation irel, Relation hrel,- LVRelStats *vacrelstats);
static void lazy_vacuum_index(Relation indrel,
IndexBulkDeleteResult **stats,
LVRelStats *vacrelstats);
@@ -742,7 +740,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats,
bool use_quick_strategy = (vacrelstats->num_dead_tuples/vacrelstats->old_live_tuples < target_index_deletion_factor);
if (use_quick_strategy && (Irel[i]->rd_amroutine->amtargetdelete != NULL))
- quick_vacuum_index(Irel[i], onerel, vacrelstats);+ quick_vacuum_index(Irel[i], onerel, vacrelstats->dead_tuples, vacrelstats->num_dead_tuples);
else
lazy_vacuum_index(Irel[i],
&indstats[i],
@@ -1394,7 +1392,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats,
bool use_quick_strategy = (vacrelstats->num_dead_tuples/vacrelstats->old_live_tuples < target_index_deletion_factor);
if (use_quick_strategy && (Irel[i]->rd_amroutine->amtargetdelete != NULL))
- quick_vacuum_index(Irel[i], onerel, vacrelstats);+ quick_vacuum_index(Irel[i], onerel, vacrelstats->dead_tuples, vacrelstats->num_dead_tuples);
else
lazy_vacuum_index(Irel[i],
&indstats[i],
@@ -1730,7 +1728,12 @@ get_tuple_by_tid(Relation rel, ItemPointer tid)
{
offnum = ItemIdGetRedirect(lp);
lp = PageGetItemId(page, offnum);
- Assert(ItemIdIsUsed(lp));++ if (!ItemIdIsUsed(lp))+ {+ UnlockReleaseBuffer(buffer);+ return NULL;+ }
}
/* Form a tuple */
@@ -1747,30 +1750,34 @@ get_tuple_by_tid(Relation rel, ItemPointer tid)
* quick_vacuum_index() -- quick vacuum one index relation.
*
* Delete all the index entries pointing to tuples listed in
- * vacrelstats->dead_tuples.+ * dead_tuples.
*/
-static void+void
quick_vacuum_index(Relation irel, Relation hrel,
- LVRelStats *vacrelstats)+ ItemPointer dead_tuples,+ int num_dead_tuples)
{
int tnum;
- bool* found = palloc0(vacrelstats->num_dead_tuples*sizeof(bool));- IndexInfo* indexInfo = BuildIndexInfo(irel);- EState* estate = CreateExecutorState();- ExprContext* econtext = GetPerTupleExprContext(estate);- ExprState* predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);- TupleTableSlot* slot = MakeSingleTupleTableSlot(RelationGetDescr(hrel));+ bool *found = palloc0(num_dead_tuples*sizeof(bool));+ IndexInfo *indexInfo = BuildIndexInfo(irel);+ EState *estate = CreateExecutorState();+ ExprContext *econtext = GetPerTupleExprContext(estate);+ ExprState *predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);+ TupleTableSlot *slot = MakeSingleTupleTableSlot(RelationGetDescr(hrel));
IndexTargetDeleteResult stats;
IndexTargetDeleteInfo ivinfo;
+ Assert(found != NULL);++ stats.tuples_removed = 0;
ivinfo.indexRelation = irel;
ivinfo.heapRelation = hrel;
econtext->ecxt_scantuple = slot;
/* Get tuple from heap */
- for (tnum = vacrelstats->num_dead_tuples-1; tnum >= 0; tnum--)+ for (tnum = num_dead_tuples-1; tnum >= 0; tnum--)
{
HeapTuple tuple;
Datum values[INDEX_MAX_KEYS];
@@ -1781,12 +1788,13 @@ quick_vacuum_index(Relation irel, Relation hrel,
continue;
/* Get a tuple from heap */
- if ((tuple = get_tuple_by_tid(hrel, &(vacrelstats->dead_tuples[tnum]))) == NULL)+ if ((tuple = get_tuple_by_tid(hrel, &(dead_tuples[tnum]))) == NULL)
{
/*
* Tuple has 'not used' status.
*/
found[tnum] = true;
+ elog(LOG, "NULL");
continue;
}
@@ -1805,6 +1813,7 @@ quick_vacuum_index(Relation irel, Relation hrel,
if ((predicate != NULL) && (!ExecQual(predicate, econtext)))
{
found[tnum] = true;
+ elog(LOG, "Predicate");
continue;
}
@@ -1814,15 +1823,16 @@ quick_vacuum_index(Relation irel, Relation hrel,
* Make attempt to delete some index entries by one tree descent.
* We use only a part of TID list, which contains not found TID's.
*/
- ivinfo.dead_tuples = vacrelstats->dead_tuples;+ ivinfo.dead_tuples = dead_tuples;
ivinfo.last_dead_tuple = tnum;
ivinfo.found_dead_tuples = found;
+
index_target_delete(&ivinfo, &stats, values, isnull);
}
- pfree(found);
ExecDropSingleTupleTableSlot(slot);
FreeExecutorState(estate);
+ pfree(found);
}
/*
diff --git a/src/backend/postmaster/Makefile b/src/backend/postmaster/Makefileindex 71c23211b2..11796375e8 100644--- a/src/backend/postmaster/Makefile+++ b/src/backend/postmaster/Makefile@@ -12,7 +12,7 @@ subdir = src/backend/postmaster
top_builddir = ../../..
include $(top_builddir)/src/Makefile.global
-OBJS = autovacuum.o bgworker.o bgwriter.o checkpointer.o fork_process.o \+OBJS = autovacuum.o bgheap.o bgworker.o bgwriter.o checkpointer.o fork_process.o \
pgarch.o pgstat.o postmaster.o startup.o syslogger.o walwriter.o
include $(top_srcdir)/src/backend/common.mk
diff --git a/src/backend/postmaster/bgheap.c b/src/backend/postmaster/bgheap.cnew file mode 100644index 0000000000..d49f294d74--- /dev/null+++ b/src/backend/postmaster/bgheap.c@@ -0,0 +1,1944 @@+/*+ * bgheap.c+ *+ * PostgreSQL integrated cleaner of HEAP and INDEX relations+ *+ * Made in autovacuum analogy. Uses 'Target' strategy for clean relations,+ * without full scan.+ * The cleaner to consist of one Launcher and Workers.+ * One worker corresponds to one database.+ * Launcher receives message from a backend {dbOid; relOid; blkno} by+ * socket and translate it to a worker by shared memory buffer. Worker receive+ * message and cleanup block of heap relation and its index relations.+ *+ * Copyright (c) 1996-2018, PostgreSQL Global Development Group+ *+ *+ * IDENTIFICATION+ * src/backend/postmaster/bgheap.c+ *+ *-------------------------------------------------------------------------+ */++#include "postgres.h"++#include <unistd.h>+#include "mb/pg_wchar.h"+#include "miscadmin.h"+#include "pgstat.h"+#include "access/genam.h"+#include "access/heapam.h"+#include "access/heapam_xlog.h"+#include "access/htup_details.h"+#include "access/nbtree.h"+#include "access/visibilitymap.h"+#include "access/xact.h"+#include "catalog/catalog.h"+#include "catalog/pg_database.h"+#include "catalog/pg_namespace.h"+#include "catalog/index.h"+#include "catalog/namespace.h"+#include "commands/progress.h"+#include "commands/vacuum.h"+#include "common/ip.h"+#include "executor/executor.h"+#include "libpq/pqsignal.h"+#include "postmaster/bgheap.h"+#include "postmaster/fork_process.h"+#include "postmaster/postmaster.h"+#include "storage/condition_variable.h"+#include "storage/fd.h"+#include "storage/freespace.h"+#include "storage/ipc.h"+#include "storage/lmgr.h"+#include "storage/pmsignal.h"+#include "storage/procarray.h"+#include "storage/smgr.h"+#include "tcop/tcopprot.h"+#include "utils/ps_status.h"+#include "utils/lsyscache.h"+#include "utils/resowner.h"+#include "utils/shash.h"+#include "utils/syscache.h"+#include "utils/timeout.h"++/*+ * Maximum number of task items in storage at a backend side before shipping to a+ * background heap cleaner+ */+#define BACKEND_DIRTY_ITEMS_MAX (20)++/*+ * Maximum number of task items in waiting list at a Launcher side.+ * It is necessary to make it sufficiently large. If a number of arrived+ * messages exceed this, we ignore the remains and increase counter of missed+ * items.+ */+#define WAITING_MESSAGES_MAX_NUM (100)++/*+ * Maximum number of dirty blocks which can keep a worker for each relation+ */+#define WORKER_DIRTYBLOCKS_MAX_NUM (1000)++/* Maximum number of slots for dirty relations */+#define WORKER_RELATIONS_MAX_NUM (100)++/* Maximum number of task items in a launcher/worker shared buffer */+#define WORKER_TASK_ITEMS_MAX (1000)++/*+ * Maximum time interval which worker can idle without a task (ms)+ */+#define WORKER_IDLE_TIME_DURATION_MAX (5000)++/* Minimal info for cleanup a block of heap relation and its index relations */+typedef struct CleanerMessage+{+ Oid dbNode; /* Database ID */+ Oid relid; /* Relation ID */+ int blkno; /* Block number */+ TransactionId xid;+ uint16 hits;+} CleanerMessage;++typedef struct WorkerTask+{+ BlockNumber blkno;+ TransactionId lastXid;+ uint16 hits;+} WorkerTask;++/*+ * Shared memory data to control and two-way communication with worker+ */+typedef struct WorkerInfoData+{+ dlist_node links;+ Oid dbOid; /* Database ID of the worker */+ TimestampTz launchtime; /* To define a time of last worker activity */+ int pid; /* Used for system signals passing */+ CleanerMessage buffer[WORKER_TASK_ITEMS_MAX]; /* Array of work items */+ int nitems; /* Number of work items in buffer */+ LWLock WorkItemLock; /* Locker for safe buffer access */+ int id; /* Used for Internal launcher buffers management */+} WorkerInfoData;++typedef struct WorkerInfoData *WorkerInfo;++/*+ * Shared memory lists.+ * Launcher get an element from freeWorkers list and init startingWorker value.+ * Worker set startingWorker to NULL value after startup and add himself+ * to runningWorkers list.+ */+typedef struct HeapCleanerShmemStruct+{+ dlist_head freeWorkers;+ dlist_head runningWorkers;+ WorkerInfo startingWorker;+} HeapCleanerShmemStruct;++typedef struct DirtyRelation+{+ Oid relid;+ SHTAB *items;+} DirtyRelation;++static int stat_vainly_cleaned_tuples = 0;+static uint64 stat_total_deletions = 0;+static int stat_false_block_hits = 0;+static uint32 stat_missed_blocks = 0;++static MemoryContext BGHeapMemCxt = NULL;++/*+ * Table of database relations.+ * For each relation we support a waiting list of dirty blocks.+ */+static PSHTAB PrivateRelationsTable = NULL;++static bool am_heapcleaner_launcher = false;+static bool am_heapcleaner_worker = false;++int heapcleaner_max_workers = 10;+static WorkerInfo MyWorkerInfo = NULL;+static HeapCleanerShmemStruct *HeapCleanerShmem;++/*+ * Hash table for collection of dirty blocks after heap_page_prune() action+ * at a Backend side.+ */+PSHTAB dblocks = NULL;++/* Signal handling */+static volatile sig_atomic_t got_SIGHUP = false;+static volatile sig_atomic_t got_SIGTERM = false;+static volatile sig_atomic_t got_SIGUSR2 = false;++NON_EXEC_STATIC pgsocket HeapCleanerSock = PGINVALID_SOCKET;+static struct sockaddr_storage HeapCleanerSockAddr;++static int TrancheId;++/*+ * Hash table for waiting data: one for each active worker and one for all messages+ * intended to not-running workers. Reasons:+ * 1. Worker in startup process.+ * 2. Another worker in startup and we can't launch new worker.+ * 3. We have not free slots for new workers.+ */+PSHTAB *wTab;++/*+ * Parameters of each hash table+ */+SHTABCTL wTabCtl;++#ifdef EXEC_BACKEND+static pid_t hclauncher_forkexec(void);+static pid_t hcworker_forkexec(void);+#endif++static uint64 CleanerMessageHashFunc(void *key, uint64 size, uint64 base);+static PSHTAB cleanup_relations(DirtyRelation *res, SHTAB *AuxiliaryList, bool got_SIGTERM);+static bool DefaultCompareFunc(void* bucket1, void* bucket2);+NON_EXEC_STATIC void HeapCleanerLauncherMain(int argc, char *argv[]) pg_attribute_noreturn();+NON_EXEC_STATIC void HeapCleanerWorkerMain(int argc, char *argv[]) pg_attribute_noreturn();+static bool isEqualMsgs(void *arg1, void *arg2);++static void launch_worker(Oid dbNode);+static WorkerInfo look_for_worker(Oid dbNode);+static void main_launcher_loop(void);+static void main_worker_loop(void);++static void SIGHUP_Handler(SIGNAL_ARGS);+static void SIGTERM_Handler(SIGNAL_ARGS);+static void SIGUSR2_Handler(SIGNAL_ARGS);+static void backend_send_dirty_blocks(void);++#ifdef EXEC_BACKEND+/*+ * forkexec routine for the autovacuum launcher process.+ *+ * Format up the arglist, then fork and exec.+ */+static pid_t+hclauncher_forkexec(void)+{+ char *av[10];+ int ac = 0;++ av[ac++] = "postgres";+ av[ac++] = "--forkhclauncher";+ av[ac++] = NULL; /* filled in by postmaster_forkexec */+ av[ac] = NULL;++ Assert(ac < lengthof(av));++ return postmaster_forkexec(ac, av);+}++/*+ * We need this set from the outside, before InitProcess is called+ */+void+HeapCleanerLauncherIAm(void)+{+ am_heapcleaner_launcher = true;+}++static pid_t+hcworker_forkexec(void)+{+ char *av[10];+ int ac = 0;++ av[ac++] = "postgres";+ av[ac++] = "--forkhcworker";+ av[ac++] = NULL; /* filled in by postmaster_forkexec */+ av[ac] = NULL;++ Assert(ac < lengthof(av));++ return postmaster_forkexec(ac, av);+}++/*+ * We need this set from the outside, before InitProcess is called+ */+void+HeapCleanerWorkerIAm(void)+{+ am_heapcleaner_worker = true;+}+#endif++/*+ * Send dirty blocks from a collector's table to launcher at the end of transaction+ */+void+AtEOXact_BGHeap_tables(bool isCommit)+{+ MemoryContext oldMemCxt;++ if (BGHeapMemCxt == NULL)+ BGHeapMemCxt = AllocSetContextCreate(BGHeapMemCxt, "bgheap",+ ALLOCSET_DEFAULT_SIZES);+ oldMemCxt = MemoryContextSwitchTo(BGHeapMemCxt);++ backend_send_dirty_blocks();++ MemoryContextSwitchTo(oldMemCxt);+}++static void+save_to_list(PSHTAB AuxiliaryList, WorkerTask *item)+{+ bool found;++ WorkerTask *new_item = (WorkerTask *)+ SHASH_Search(AuxiliaryList,+ (void *) &(item->blkno),+ HASH_ENTER, &found);++ Assert((new_item != NULL) && (!found));+ new_item->hits = item->hits;+ new_item->lastXid = item->lastXid;+}++/*+ * Main logic of HEAP and index relations cleaning+ */+static PSHTAB+cleanup_relations(DirtyRelation *res, PSHTAB AuxiliaryList, bool got_SIGTERM)+{+ Relation heapRelation;+ Relation *IndexRelations;+ int nindexes;+ LOCKMODE lockmode = ExclusiveLock/*AccessShareLock*/;+ WorkerTask *item;++ Assert(res != NULL);+ Assert(res->items != NULL);+ Assert(AuxiliaryList != NULL);+ Assert(SHASH_Entries(AuxiliaryList) == 0);++ if (SHASH_Entries(res->items) == 0)+ return AuxiliaryList;++ if (RecoveryInProgress())+ {+ SHASH_Clean(res->items);+ return AuxiliaryList;+ }++ CHECK_FOR_INTERRUPTS();++ StartTransactionCommand();++ /*+ * At this point relation availability is not guaranteed.+ * Make safe test to check this.+ */+ heapRelation = try_relation_open(res->relid, lockmode);++ if (!heapRelation)+ {+ elog(LOG, "[%d] Cleanup: UnSuccessful opening. Relation deleted? dbOid=%d", res->relid, MyWorkerInfo->dbOid);+ CommitTransactionCommand();+ SHASH_Clean(res->items);+ return AuxiliaryList;+ }++ /*+ * Check relation type similarly vacuum+ */+ if (!(pg_class_ownercheck(RelationGetRelid(heapRelation), GetUserId()) ||+ (pg_database_ownercheck(MyDatabaseId, GetUserId()) && !heapRelation->rd_rel->relisshared)))+ {+ if (heapRelation->rd_rel->relisshared)+ ereport(WARNING,+ (errmsg("skipping \"%s\" --- only superuser can vacuum it",+ RelationGetRelationName(heapRelation))));+ else if (heapRelation->rd_rel->relnamespace == PG_CATALOG_NAMESPACE)+ ereport(WARNING,+ (errmsg("skipping \"%s\" --- only superuser or database owner can vacuum it",+ RelationGetRelationName(heapRelation))));+ else+ ereport(WARNING,+ (errmsg("skipping \"%s\" --- only table or database owner can vacuum it",+ RelationGetRelationName(heapRelation))));+ relation_close(heapRelation, lockmode);+ CommitTransactionCommand();+ SHASH_Clean(res->items);+ return AuxiliaryList;+ }++ /* Open and lock index relations correspond to the heap relation */+ vac_open_indexes(heapRelation, RowExclusiveLock, &nindexes, &IndexRelations);++ /* Main cleanup cycle */+ for (SHASH_SeqReset(res->items);+ (item = (WorkerTask *) SHASH_SeqNext(res->items)) != NULL; )+ {+ BlockNumber nblocks;+ bool needLock;+ Buffer buffer;+ ItemPointerData dead_tuples[MaxOffsetNumber];+ int num_dead_tuples = 0;+ OffsetNumber offnum;+ int irnum;+ Page page;+ ItemId lp;+ int tnum;+ bool found_non_nbtree = false;++ Assert(item->hits > 0);++ if (!got_SIGTERM &&+ (res->items->Header.ElementsMaxNum/SHASH_Entries(res->items) > 3) &&+ (item->lastXid > GetOldestXmin(NULL, PROCARRAY_FLAGS_VACUUM)) &&+ (item->hits < 10))+ {+ /*+ * Skip block cleaning and save it to a waiting list+ * for the next iteration.+ */+ save_to_list(AuxiliaryList, item);+ continue;+ }++ needLock = !RELATION_IS_LOCAL(heapRelation);+ if (needLock)+ LockRelationForExtension(heapRelation, ExclusiveLock);+ nblocks = RelationGetNumberOfBlocks(heapRelation);+ if (needLock)+ UnlockRelationForExtension(heapRelation, ExclusiveLock);++ if (item->blkno >= nblocks)+ {+ /*+ * Block was deleted early.+ * Skip cleaning and drop block from waiting list.+ * Accumulate some stats before continue.+ */+ stat_false_block_hits += item->hits;+ pgstat_progress_update_param(PROGRESS_CLEANER_FALSE_HITS, stat_false_block_hits);+ continue;+ }++ /* Create TID list */+ if (!got_SIGTERM)+ buffer = ReadBufferExtended(heapRelation, MAIN_FORKNUM, item->blkno, RBM_NORMAL_NO_READ, NULL);+ else+ buffer = ReadBuffer(heapRelation, item->blkno);++ if (BufferIsInvalid(buffer))+ {+ /*+ * Buffer was already evicted from shared buffers+ */+ Assert(!got_SIGTERM);+ save_to_list(AuxiliaryList, item);+ continue;+ }++ if (!ConditionalLockBuffer(buffer))+ {+ /* Can't lock buffer. */+ ReleaseBuffer(buffer);+ save_to_list(AuxiliaryList, item);+ continue;+ }++ if (!IsBufferDirty(buffer))+ {+ /* Skip block if it is not dirty */+ UnlockReleaseBuffer(buffer);+ continue;+ }++ page = BufferGetPage(buffer);++ /* Collect dead tuples TID's */+ for (offnum = FirstOffsetNumber;+ offnum < PageGetMaxOffsetNumber(page);+ offnum = OffsetNumberNext(offnum))+ {+ lp = PageGetItemId(page, offnum);+ if (ItemIdIsDead(lp)/* && ItemIdHasStorage(lp)*/)+ {+ ItemPointerSet(&(dead_tuples[num_dead_tuples]), item->blkno, offnum);+ num_dead_tuples++;+ }+ }+ LockBuffer(buffer, BUFFER_LOCK_UNLOCK);++ if (num_dead_tuples == 0)+ {+ /*+ * Block is clear. Accumulate some stats and continue.+ */+ stat_vainly_cleaned_tuples += item->hits;+ pgstat_progress_update_param(PROGRESS_CLEANER_VAINLY_CLEANED_TUPLES, stat_vainly_cleaned_tuples);+ ReleaseBuffer(buffer);+ continue;+ }++ /* Iterate across all index relations */+ for (irnum = 0; irnum < nindexes; irnum++)+ {+ if (IndexRelations[irnum]->rd_amroutine->amtargetdelete == NULL)+ {+ /*+ * Can clean only btree indexes now.+ */+ found_non_nbtree = true;+ continue;+ }++ quick_vacuum_index(IndexRelations[irnum],+ heapRelation,+ dead_tuples,+ num_dead_tuples);+ }++ /*+ * If heap relation has not only b-tree indexes, can't clean heap block.+ */+ if (!found_non_nbtree)+ {+ OffsetNumber unusable[MaxOffsetNumber];+ int nunusable = 0;+ Size freespace;++ if (!ConditionalLockBufferForCleanup(buffer))+ {+ ReleaseBuffer(buffer);+ save_to_list(AuxiliaryList, item);+ continue;+ }++ START_CRIT_SECTION();++ /* Release DEAD heap tuples storage */+ for (tnum = 0; tnum < num_dead_tuples; tnum++)+ {+ OffsetNumber offnum = ItemPointerGetOffsetNumber(&dead_tuples[tnum]);+ ItemId lp = PageGetItemId(page, offnum);++ Assert(ItemIdIsDead(lp));+ ItemIdSetUnused(lp);+ unusable[nunusable++] = offnum;+ }++ if (nunusable > 0)+ {+ PageRepairFragmentation(page);+ PageClearFull(page);+ MarkBufferDirty(buffer);++ if (RelationNeedsWAL(heapRelation))+ {+ XLogRecPtr recptr = log_heap_clean(heapRelation, buffer,+ NULL, 0,+ NULL, 0,+ unusable, nunusable,+ InvalidTransactionId);++ PageSetLSN(BufferGetPage(buffer), recptr);+ }+ stat_total_deletions += nunusable;+ pgstat_progress_update_param(PROGRESS_CLEANER_TOTAL_DELETIONS, stat_total_deletions);+ }++ END_CRIT_SECTION();+ freespace = PageGetHeapFreeSpace(page);++ UnlockReleaseBuffer(buffer);+ RecordPageWithFreeSpace(heapRelation, item->blkno, freespace);+ pgstat_update_heap_dead_tuples(heapRelation, nunusable);+ }+ /*+ * ToDo: that we will do with TOAST relation?+ */+ }++ vac_close_indexes(nindexes, IndexRelations, RowExclusiveLock/*NoLock*/);+ relation_close(heapRelation, lockmode);+ CommitTransactionCommand();++ {+ SHTAB *tmp = res->items;+ SHASH_Clean(tmp);+ res->items = AuxiliaryList;+ return tmp;+ }+}++static bool+DefaultCompareFunc(void* bucket1, void* bucket2)+{+ WorkerTask *item1 = (WorkerTask *) bucket1;+ WorkerTask *item2 = (WorkerTask *) bucket2;++ if (item1->blkno == item2->blkno)+ return true;+ else+ return false;+}++/*+ * Free any launcher resources before close process+ */+static void+FreeLauncherInfo(int code, Datum arg)+{+ int counter;++ for (counter = 0; counter < heapcleaner_max_workers+1; counter++)+ SHASH_Destroy(wTab[counter]);+ pfree(wTab);+}++/*+ * Return a WorkerInfo to the free list+ */+static void+FreeWorkerInfo(int code, Datum arg)+{+ if (MyWorkerInfo != NULL)+ {+ LWLockAcquire(HeapCleanerLock, LW_EXCLUSIVE);++ dlist_delete(&MyWorkerInfo->links);+ MyWorkerInfo->dbOid = InvalidOid;+ MyWorkerInfo->launchtime = 0;+ dlist_push_head(&HeapCleanerShmem->freeWorkers,+ &MyWorkerInfo->links);+ /* not mine anymore */+ MyWorkerInfo = NULL;++ LWLockRelease(HeapCleanerLock);+ }+ else+ elog(ERROR, "---> MyWorkerInfo is NULL");+}++/*+ * Initialize communication with backends+ */+void+HeapCleanerInit()+{+ struct addrinfo *addrs = NULL,+ *addr, hints;+ ACCEPT_TYPE_ARG3 alen;+ int ret;+ fd_set rset;+ struct timeval tv;+ char test_byte;+ int sel_res;+ int tries = 0;++#define TESTBYTEVAL ((char) 199)++ /*+ * Create the UDP socket for sending and receiving statistic messages+ */+ hints.ai_flags = AI_PASSIVE;+ hints.ai_family = AF_UNSPEC;+ hints.ai_socktype = SOCK_DGRAM;+ hints.ai_protocol = 0;+ hints.ai_addrlen = 0;+ hints.ai_addr = NULL;+ hints.ai_canonname = NULL;+ hints.ai_next = NULL;+ ret = pg_getaddrinfo_all("localhost", NULL, &hints, &addrs);+ Assert(ret == 0);++ for (addr = addrs; addr; addr = addr->ai_next)+ {+#ifdef HAVE_UNIX_SOCKETS+ /* Ignore AF_UNIX sockets, if any are returned. */+ if (addr->ai_family == AF_UNIX)+ continue;+#endif++ if (++tries > 1)+ ereport(LOG,+ (errmsg("trying another address for the statistics collector")));++ if ((HeapCleanerSock = socket(addr->ai_family, SOCK_DGRAM, 0)) == PGINVALID_SOCKET)+ {+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not create socket for relation cleaner: %m")));+ continue;+ }++ if (bind(HeapCleanerSock, addr->ai_addr, addr->ai_addrlen) < 0)+ {+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not bind socket for relation cleaner: %m")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }++ alen = sizeof(HeapCleanerSockAddr);+ if (getsockname(HeapCleanerSock, (struct sockaddr *) &HeapCleanerSockAddr, &alen) < 0)+ {+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not get address of socket for statistics collector: %m")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }++ if (connect(HeapCleanerSock, (struct sockaddr *) &HeapCleanerSockAddr, alen) < 0)+ {+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not connect socket for statistics collector: %m")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }+ /*+ * Try to send and receive a one-byte test message on the socket. This+ * is to catch situations where the socket can be created but will not+ * actually pass data (for instance, because kernel packet filtering+ * rules prevent it).+ */+ test_byte = TESTBYTEVAL;++retry1:+ if (send(HeapCleanerSock, &test_byte, 1, 0) != 1)+ {+ if (errno == EINTR)+ goto retry1; /* if interrupted, just retry */+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not send test message on socket for statistics collector: %m")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }++ /*+ * There could possibly be a little delay before the message can be+ * received. We arbitrarily allow up to half a second before deciding+ * it's broken.+ */+ for (;;) /* need a loop to handle EINTR */+ {+ FD_ZERO(&rset);+ FD_SET(HeapCleanerSock, &rset);++ tv.tv_sec = 0;+ tv.tv_usec = 500000;+ sel_res = select(HeapCleanerSock + 1, &rset, NULL, NULL, &tv);+ if (sel_res >= 0 || errno != EINTR)+ break;+ }+ if (sel_res < 0)+ {+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("select() failed in statistics collector: %m")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }+ if (sel_res == 0 || !FD_ISSET(HeapCleanerSock, &rset))+ {+ /*+ * This is the case we actually think is likely, so take pains to+ * give a specific message for it.+ *+ * errno will not be set meaningfully here, so don't use it.+ */+ ereport(LOG,+ (errcode(ERRCODE_CONNECTION_FAILURE),+ errmsg("test message did not get through on socket for statistics collector")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }++ test_byte++; /* just make sure variable is changed */++retry2:+ if (recv(HeapCleanerSock, &test_byte, 1, 0) != 1)+ {+ if (errno == EINTR)+ goto retry2; /* if interrupted, just retry */+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not receive test message on socket for statistics collector: %m")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }++ if (test_byte != TESTBYTEVAL) /* strictly paranoia ... */+ {+ ereport(LOG,+ (errcode(ERRCODE_INTERNAL_ERROR),+ errmsg("incorrect test message transmission on socket for statistics collector")));+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+ continue;+ }++ /* If we get here, we have a working socket */++ break;+ }++ /* Did we find a working address? */+ if (!addr || HeapCleanerSock == PGINVALID_SOCKET)+ goto startup_failed;++ /*+ * Set the socket to non-blocking IO. This ensures that if the collector+ * falls behind, statistics messages will be discarded; backends won't+ * block waiting to send messages to the collector.+ */+ if (!pg_set_noblock(HeapCleanerSock))+ {+ ereport(LOG,+ (errcode_for_socket_access(),+ errmsg("could not set statistics collector socket to nonblocking mode: %m")));+ goto startup_failed;+ }+ pg_freeaddrinfo_all(hints.ai_family, addrs);++ return;++startup_failed:+ ereport(LOG,+ (errmsg("disabling heap cleaner for lack of working socket")));++ if (addrs)+ pg_freeaddrinfo_all(hints.ai_family, addrs);++ if (HeapCleanerSock != PGINVALID_SOCKET)+ closesocket(HeapCleanerSock);+ HeapCleanerSock = PGINVALID_SOCKET;+}++/*+ * Start and initialization logic of a launcher+ */+NON_EXEC_STATIC void+HeapCleanerLauncherMain(int argc, char *argv[])+{+ sigjmp_buf local_sigjmp_buf;+ MemoryContext bgheap_context;+ int counter;++ am_heapcleaner_launcher = true;++ /*+ * Identify myself via ps+ */+ init_ps_display(pgstat_get_backend_desc(B_BG_HEAPCLNR_LAUNCHER), "", "", "");++ SetProcessingMode(InitProcessing);++ pqsignal(SIGHUP, SIGHUP_Handler); /* set flag to read config file */+ pqsignal(SIGINT, StatementCancelHandler);+ pqsignal(SIGTERM, SIGTERM_Handler);+ pqsignal(SIGQUIT, quickdie); /* hard crash time */+ InitializeTimeouts();+ pqsignal(SIGPIPE, SIG_IGN);+ pqsignal(SIGUSR1, procsignal_sigusr1_handler);+ pqsignal(SIGUSR2, SIG_IGN);+ pqsignal(SIGFPE, FloatExceptionHandler);+ pqsignal(SIGCHLD, SIG_DFL);++ BaseInit();++#ifndef EXEC_BACKEND+ InitProcess();+#endif++ InitPostgres(NULL, InvalidOid, NULL, InvalidOid, NULL, false);++ SetProcessingMode(NormalProcessing);++ /*+ * Create a resource owner to keep track of our resources (currently only+ * buffer pins).+ */+ CurrentResourceOwner = ResourceOwnerCreate(NULL, "Background cleaner");++ bgheap_context = AllocSetContextCreate(TopMemoryContext,+ "Background cleaner",+ ALLOCSET_DEFAULT_SIZES);+ MemoryContextSwitchTo(bgheap_context);++ if (sigsetjmp(local_sigjmp_buf, 1) != 0)+ {+ /* Since not using PG_TRY, must reset error stack by hand */+ error_context_stack = NULL;++ /* Prevent interrupts while cleaning up */+ HOLD_INTERRUPTS();++ /* Forget any pending QueryCancel or timeout request */+ disable_all_timeouts(false);+ /* Report the error to the server log */+ EmitErrorReport();++ LWLockReleaseAll();+ pgstat_report_wait_end();+ AbortBufferIO();+ UnlockBuffers();+ if (CurrentResourceOwner)+ /* buffer pins are released here: */+ ResourceOwnerRelease(CurrentResourceOwner,+ RESOURCE_RELEASE_BEFORE_LOCKS,+ false, true);+ /* we needn't bother with the other ResourceOwnerRelease phases */+ AtEOXact_Buffers(false);+ AtEOXact_SMgr();+ AtEOXact_Files(false);+ AtEOXact_HashTables(false);++ /*+ * Now return to normal top-level context and clear ErrorContext for+ * next time.+ */+ MemoryContextSwitchTo(bgheap_context);+ FlushErrorState();++ /* Flush any leaked data in the top-level context */+ MemoryContextResetAndDeleteChildren(bgheap_context);++ /*+ * Make sure pgstat also considers our stat data as gone. Note: we+ * mustn't use autovac_refresh_stats here.+ */+ pgstat_clear_snapshot();++ /* Now we can allow interrupts again */+ RESUME_INTERRUPTS();++ if (got_SIGTERM)+ proc_exit(0);+ /*+ * Sleep at least 1 second after any error. A write error is likely+ * to be repeated, and we don't want to be filling the error logs as+ * fast as we can.+ */+ pg_usleep(1000000L);+ }++ /* We can now handle ereport(ERROR) */+ PG_exception_stack = &local_sigjmp_buf;++ /*+ * Unblock signals (they were blocked when the postmaster forked us)+ */+ PG_SETMASK(&UnBlockSig);++ wTabCtl.FillFactor = 0.75;+ wTabCtl.ElementsMaxNum = WAITING_MESSAGES_MAX_NUM;+ wTabCtl.ElementSize = sizeof(CleanerMessage);+ wTabCtl.KeySize = 2 * sizeof(Oid) + sizeof(BlockNumber);+ wTabCtl.HashFunc = CleanerMessageHashFunc;+ wTabCtl.CompFunc = isEqualMsgs;++ Assert(heapcleaner_max_workers == 10);+ wTab = palloc0((heapcleaner_max_workers+1)*sizeof(PSHTAB));+ for (counter = 0; counter < heapcleaner_max_workers+1; counter++)+ wTab[counter] = SHASH_Create(wTabCtl);+ on_shmem_exit(FreeLauncherInfo, 0);+elog(LOG, "-> Launcher Started!");+ main_launcher_loop();++ proc_exit(0);+}++static bool+backend_store_dirty_block(CleanerMessage *msg)+{+ bool found;+ CleanerMessage *rec;++ /* Initialize collector's hash table */+ if (dblocks == NULL)+ {+ SHTABCTL ctl;++ ctl.ElementsMaxNum = BACKEND_DIRTY_ITEMS_MAX;+ ctl.FillFactor = 0.75;+ ctl.ElementSize = sizeof(CleanerMessage);+ ctl.KeySize = 2 * sizeof(Oid) + sizeof(BlockNumber);+ ctl.HashFunc = CleanerMessageHashFunc;+ ctl.CompFunc = isEqualMsgs;+ dblocks = SHASH_Create(ctl);+ }++ /* Search for new position or for duplicates */+ rec = (CleanerMessage *) SHASH_Search(dblocks, (void *) msg, SHASH_ENTER, &found);++ if (!rec)+ return true;++ if (!found)+ rec->hits = 0;++ rec->xid = msg->xid;+ rec->hits++;+ return false;+}++/*+ * Send a package of dirty blocks from a Backend to the Cleaner.+ * Form the package by passing across the table. During this pass table is+ * cleaned. Sign of clean row is (dbNode == 0) condition.+ */+static void+backend_send_dirty_blocks(void)+{+ int rc;+ CleanerMessage *rec;+ CleanerMessage data[BACKEND_DIRTY_ITEMS_MAX];+ int nitems = 0;++ if (dblocks == NULL)+ return;++ if (SHASH_Entries(dblocks) == 0)+ return;++ Assert(SHASH_Entries(dblocks) <= BACKEND_DIRTY_ITEMS_MAX);++ for (SHASH_SeqReset(dblocks);+ (rec = SHASH_SeqNext(dblocks)) != NULL; )+ {+ Assert(rec->hits > 0);+ Assert(nitems < BACKEND_DIRTY_ITEMS_MAX);+ data[nitems++] = *rec;+ }+ SHASH_Clean(dblocks);++ /* Send data to launcher.+ * We'll retry after EINTR, but ignore all other failures+ */+ do+ {+ rc = send(HeapCleanerSock, data, nitems*sizeof(CleanerMessage), 0);+ } while (rc < 0 && errno == EINTR);+}++/*+ * Send a dirty block of relation to the Cleaner.+ * Before shipping to the Cleaner dirty blocks collects in a simple hash+ * table. Duplicates are discarded.+ */+void+HeapCleanerSend(Relation relation, BlockNumber blkno)+{+ CleanerMessage msg;+ MemoryContext oldMemCxt;++ if (BGHeapMemCxt == NULL)+ BGHeapMemCxt = AllocSetContextCreate(BGHeapMemCxt, "bgheap",+ ALLOCSET_DEFAULT_SIZES);++ if (RecoveryInProgress())+ return;++ if (HeapCleanerSock == PGINVALID_SOCKET)+ return;++ /*+ * ToDo: Cleaning of system relations make instability in regression tests+ * output order+ */+ if (IsSystemRelation(relation))+ return;++ /*+ * Check the relation type.+ * Not all relations can be cleaned now (similarly vacuum).+ */+ if (relation->rd_rel->relkind != RELKIND_RELATION &&+ relation->rd_rel->relkind != RELKIND_MATVIEW &&+ relation->rd_rel->relkind != RELKIND_TOASTVALUE &&+ relation->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)+ {+ ereport(WARNING,+ (errmsg("skipping \"%s\" --- cannot clean non-tables or special system tables",+ RelationGetRelationName(relation))));+ return;+ }++ if (RELATION_IS_OTHER_TEMP(relation))+ return;++ if (relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)+ return;++ oldMemCxt = MemoryContextSwitchTo(BGHeapMemCxt);++ msg.dbNode = relation->rd_node.dbNode;+ msg.relid = RelationGetRelid(relation);+ msg.blkno = blkno;+ msg.xid = GetCurrentTransactionIdIfAny();++ if (backend_store_dirty_block(&msg))+ backend_send_dirty_blocks();++ MemoryContextSwitchTo(oldMemCxt);+}++/*+ * Start and initialization logic of a worker+ */+NON_EXEC_STATIC void+HeapCleanerWorkerMain(int argc, char *argv[])+{+ sigjmp_buf local_sigjmp_buf;++ am_heapcleaner_worker = true;++ /* Identify myself via ps */+ init_ps_display(pgstat_get_backend_desc(B_BG_HEAPCLNR_WORKER), "", "", "");++ SetProcessingMode(InitProcessing);++ pqsignal(SIGHUP, SIGHUP_Handler);+ pqsignal(SIGINT, StatementCancelHandler);+ pqsignal(SIGTERM, SIGTERM_Handler);+ pqsignal(SIGQUIT, quickdie);+ InitializeTimeouts(); /* establishes SIGALRM handler */++ pqsignal(SIGPIPE, SIG_IGN);+ pqsignal(SIGUSR1, procsignal_sigusr1_handler);+ pqsignal(SIGUSR2, SIGUSR2_Handler); /* Signal: Buffer contains a message */+ pqsignal(SIGFPE, FloatExceptionHandler);+ pqsignal(SIGCHLD, SIG_DFL);++ /* Early initialization */+ BaseInit();++#ifndef EXEC_BACKEND+ InitProcess();+#endif++ if (sigsetjmp(local_sigjmp_buf, 1) != 0)+ {+ /* Prevents interrupts while cleaning up */+ HOLD_INTERRUPTS();++ /* Report the error to the server log */+ EmitErrorReport();++ proc_exit(0);+ }++ /* We can now handle ereport(ERROR) */+ PG_exception_stack = &local_sigjmp_buf;++ PG_SETMASK(&UnBlockSig);++ /* Initialization steps */+ LWLockAcquire(HeapCleanerLock, LW_EXCLUSIVE);+ if (HeapCleanerShmem->startingWorker != NULL)+ {+ MyWorkerInfo = HeapCleanerShmem->startingWorker;+ MyWorkerInfo->pid = getpid();+ HeapCleanerShmem->startingWorker = NULL;++ /* insert into the running list */+ dlist_push_head(&HeapCleanerShmem->runningWorkers, &MyWorkerInfo->links);++ on_shmem_exit(FreeWorkerInfo, 0);+ }+ else+ elog(ERROR, "No Starting worker!");+ LWLockRelease(HeapCleanerLock);++ if (OidIsValid(MyWorkerInfo->dbOid))+ {+ char dbname[NAMEDATALEN];+ SHTABCTL pr_ctl;++ pgstat_report_heapcleaner(MyWorkerInfo->dbOid);+ InitPostgres(NULL, MyWorkerInfo->dbOid, NULL, InvalidOid, dbname, false);+ SetProcessingMode(NormalProcessing);+ set_ps_display(dbname, false);++ pr_ctl.ElementsMaxNum = 1000;+ pr_ctl.FillFactor = 0.5;+ pr_ctl.HashFunc = DefaultHashValueFunc;+ pr_ctl.CompFunc = DefaultCompareFunc;+ pr_ctl.ElementSize = sizeof(DirtyRelation);+ pr_ctl.KeySize = sizeof(int32);++ PrivateRelationsTable = SHASH_Create(pr_ctl);++ /* Add tracking info to pgstat */+ pgstat_progress_start_command(PROGRESS_COMMAND_CLEANER, MyWorkerInfo->dbOid);++ main_worker_loop();+ }+ else+ elog(ERROR, "dbid %d not valid!", MyWorkerInfo->dbOid);++ proc_exit(0);+}++/*+ * Compare messages+ */+static bool+isEqualMsgs(void *arg1, void *arg2)+{+ CleanerMessage *msg1 = arg1;+ CleanerMessage *msg2 = arg2;++ Assert(arg1 != NULL);+ Assert(arg2 != NULL);++ if (msg1->blkno != msg2->blkno)+ return false;+ if (msg1->relid != msg2->relid)+ return false;+ if (msg1->dbNode != msg2->dbNode)+ return false;++ return true;+}++/*+ * IsHeapCleaner functions+ * Return whether this is either a launcher heap cleaner process or+ * a worker process.+ */+bool+IsHeapCleanerLauncherProcess(void)+{+ return am_heapcleaner_launcher;+}++bool+IsHeapCleanerWorkerProcess(void)+{+ return am_heapcleaner_worker;+}++/*+ * Send signal to Postmaster for launch new worker instance.+ * Eject one node from freeWorkers list and assign to startingWorker+ * HeapCleanerLock must be exclusive-locked+ */+static void+launch_worker(Oid dbNode)+{+ WorkerInfo worker;+ dlist_node *wptr;++ if (dlist_is_empty(&HeapCleanerShmem->freeWorkers))+ elog(ERROR, "NO a free slot for background cleaner worker!");++ wptr = dlist_pop_head_node(&HeapCleanerShmem->freeWorkers);+ worker = dlist_container(WorkerInfoData, links, wptr);+ worker->dbOid = dbNode;+ worker->launchtime = GetCurrentTimestamp();+ worker->nitems = 0;+ HeapCleanerShmem->startingWorker = worker;++ SendPostmasterSignal(PMSIGNAL_START_HEAPCLNR_WORKER);+}++/*+ * Return worker that initialized for dbNode database or NULL+ * Caller need to acquire share lock on HeapCleanerLock+ */+static WorkerInfo+look_for_worker(Oid dbNode)+{+ dlist_node *node;+ WorkerInfo worker = NULL;++ if (!dlist_is_empty(&HeapCleanerShmem->runningWorkers))+ for (node = dlist_head_node(&HeapCleanerShmem->runningWorkers);+ ;+ node = dlist_next_node(&HeapCleanerShmem->runningWorkers, node))+ {+ if (((WorkerInfo)node)->dbOid == dbNode)+ {+ worker = (WorkerInfo) node;+ break;+ }+ if (!dlist_has_next(&HeapCleanerShmem->runningWorkers, node))+ break;+ }++ return worker;+}++/*+ * Entry point of a launcher behavior logic+ */+static void+main_launcher_loop()+{+ bool found;+ long timeout = -1L;++ Assert(HeapCleanerSock != PGINVALID_SOCKET);++ /*+ * Do work until not catch a SIGTERM signal and all blocks+ * not shipped to workers.+ * timeout <= 0 tells us that the launcher does not have any work right now,+ * and it will sleep until the socket message or the system signal.+ */+ while (!got_SIGTERM || (timeout > 0))+ {+ int rc;+ int len;+ CleanerMessage *msg;+ CleanerMessage table[BACKEND_DIRTY_ITEMS_MAX];+ WorkerInfo startingWorker;+ WorkerInfo worker;+ dlist_node *node;++ timeout = -1L;++ /* Process system signals */+ if (got_SIGHUP)+ {+ got_SIGHUP = false;+ ProcessConfigFile(PGC_SIGHUP);+ }++ ResetLatch(MyLatch);++ /* At First, receive a message from backend */+ len = recv(HeapCleanerSock, table, BACKEND_DIRTY_ITEMS_MAX*sizeof(CleanerMessage), 0);++ LWLockAcquire(HeapCleanerLock, LW_EXCLUSIVE);+ startingWorker = HeapCleanerShmem->startingWorker;++ /*+ * Message was received from socket.+ */+ if (len > 0)+ {+ CleanerMessage *mptr;++ if (len%sizeof(CleanerMessage) != 0)+ elog(ERROR, "INCORRECT Message size!");++ /*+ * Push all messages received from a backend to general+ * hash table with waiting messages.+ */+ for (mptr = table; ((char *)mptr-(char *)table) < len; mptr++)+ {+ worker = look_for_worker(mptr->dbNode);+ if (worker)+ msg = (CleanerMessage *) SHASH_Search(wTab[worker->id], mptr, SHASH_ENTER, &found);+ else+ msg = (CleanerMessage *) SHASH_Search(wTab[heapcleaner_max_workers], mptr, SHASH_ENTER, &found);++ if (msg == NULL)+ /*+ * Hash table is FULL and we can't insert a message+ */+ stat_missed_blocks++;+ else if (found)+ {+ msg->hits++;+ msg->xid = (msg->xid > mptr->xid) ? msg->xid : mptr->xid;+ }+ else+ memcpy(msg, mptr, sizeof(CleanerMessage));+ }+ }++ /*+ * Main waiting list parsing.+ */+ if (SHASH_Entries(wTab[heapcleaner_max_workers]) > 0)+ {+ bool startWorker = (startingWorker != NULL);++ /*+ * Pass across general waiting list.+ * Try to send tasks to active workers.+ */+ for (SHASH_SeqReset(wTab[heapcleaner_max_workers]);+ (msg = (CleanerMessage *) SHASH_SeqNext(wTab[heapcleaner_max_workers])) != NULL; )+ {+ worker = look_for_worker(msg->dbNode);++ if (worker)+ {+ CleanerMessage *temp_msg = (CleanerMessage *) SHASH_Search(wTab[worker->id], (void *) msg, SHASH_ENTER, &found);+ if (temp_msg != NULL)+ if (found)+ {+ temp_msg->hits += msg->hits;+ temp_msg->xid = (msg->xid > temp_msg->xid) ? msg->xid : temp_msg->xid;+ }+ else+ memcpy(temp_msg, msg, sizeof(CleanerMessage));+ else+ stat_missed_blocks++;++ /*+ * Message gone to worker. delete from main waiting list.+ */+ temp_msg = (CleanerMessage *) SHASH_Search(wTab[heapcleaner_max_workers], (void *) msg, SHASH_REMOVE, NULL);+ Assert(temp_msg != NULL);++ continue;+ }+ else if (!startWorker)+ {+ /* Start new worker */+ launch_worker(msg->dbNode);+ startWorker = true;+ }+ }+ }++ /*+ * Check: if we can't process all incoming messages, we need too small+ * timeout, check latches and go to next iteration.+ */+ if (SHASH_Entries(wTab[heapcleaner_max_workers]) > 0)+ timeout = 1L;++ /*+ * See waiting lists of active workers and try to send messages.+ */+ if (!dlist_is_empty(&HeapCleanerShmem->runningWorkers))+ {+ for (node = dlist_head_node(&HeapCleanerShmem->runningWorkers);+ ;+ node = dlist_next_node(&HeapCleanerShmem->runningWorkers, node))+ {+ worker = (WorkerInfo) node;++ if (SHASH_Entries(wTab[worker->id]) == 0)+ {+ /* Check worker idle time */+ if (TimestampDifferenceExceeds(worker->launchtime, GetCurrentTimestamp(), WORKER_IDLE_TIME_DURATION_MAX))+ {+ LWLockAcquire(&worker->WorkItemLock, LW_EXCLUSIVE);++ /*+ * Check: may be worker has tasks but is too lazy+ */+ if (worker->nitems == 0)+ {+ /* Shutdown the idle worker */+ kill(worker->pid, SIGTERM);+ dlist_delete(&worker->links);+ dlist_push_head(&HeapCleanerShmem->freeWorkers, &worker->links);++ if (!dlist_is_empty(&HeapCleanerShmem->runningWorkers))+ node = dlist_head_node(&HeapCleanerShmem->runningWorkers);+ else+ {+ LWLockRelease(&worker->WorkItemLock);+ break;+ }+ }+ LWLockRelease(&worker->WorkItemLock);+ }+ else if (!dlist_has_next(&HeapCleanerShmem->runningWorkers, node))+ break;+ else+ continue;+ }++ /* Put list of potentially dirty blocks to the worker shared buffer */+ LWLockAcquire(&worker->WorkItemLock, LW_EXCLUSIVE);++ SHASH_SeqReset(wTab[worker->id]);+ while (((msg = (CleanerMessage *) SHASH_SeqNext(wTab[worker->id])) != NULL) &&+ (worker->nitems < WORKER_TASK_ITEMS_MAX))+ {+ void *temp_msg;++ memcpy(&worker->buffer[worker->nitems++], msg, sizeof(CleanerMessage));+ temp_msg = SHASH_Search(wTab[worker->id], (void *) msg, SHASH_REMOVE, NULL);+ Assert(temp_msg != NULL);+ Assert(worker->buffer[worker->nitems-1].hits > 0);+ }++ worker->launchtime = GetCurrentTimestamp();+ LWLockRelease(&worker->WorkItemLock);++ /* Worker, you have a task! */+ kill(worker->pid, SIGUSR2);++ /*+ * If launcher has any tasks, It check signals quickly+ * and to work on further.+ */+ if (SHASH_Entries(wTab[worker->id]) > 0)+ timeout = 5L;++ if (!dlist_has_next(&HeapCleanerShmem->runningWorkers, node))+ break;+ }++ /*+ * Launcher have'nt any immediate tasks and+ * need to manage idle workers only+ */+ if (timeout < 0)+ /* We only need to wait idle workers */+ timeout = 100L;+ }+ LWLockRelease(HeapCleanerLock);++ if (!got_SIGTERM)+ {+ int wakeEvents = WL_LATCH_SET | WL_POSTMASTER_DEATH | WL_SOCKET_READABLE;++ if (timeout > 0)+ wakeEvents |= WL_TIMEOUT;+ /* Wait data or signals */+ rc = WaitLatchOrSocket(MyLatch,+ wakeEvents,+ HeapCleanerSock, timeout,+ WAIT_EVENT_BGHEAP_MAIN);+ }++ /* Emergency bailout if postmaster has died */+ if (rc & WL_POSTMASTER_DEATH)+ {+ elog(LOG, "Heap Launcher exit with 1");+ proc_exit(1);+ }+ }++ elog(LOG, "Heap Launcher exit with 0");+ proc_exit(0);+}++/*+ * Entry point of a worker behavior logic+ */+static void+main_worker_loop(void)+{+ CleanerMessage task_item[WORKER_TASK_ITEMS_MAX];+ DirtyRelation *dirty_relation[WORKER_RELATIONS_MAX_NUM];+ PSHTAB FreeDirtyBlocksList;+ long timeout = -1L;+ int dirty_relations_num = 0;+ int task_items_num = 0;+ SHTABCTL shctl;++ shctl.FillFactor = 0.75;+ shctl.ElementsMaxNum = WORKER_DIRTYBLOCKS_MAX_NUM;+ shctl.ElementSize = sizeof(WorkerTask);+ shctl.KeySize = sizeof(BlockNumber);+ shctl.HashFunc = DefaultHashValueFunc;+ shctl.CompFunc = DefaultCompareFunc;++ FreeDirtyBlocksList = SHASH_Create(shctl);++ while (!got_SIGTERM || (timeout > 0))+ {+ int rc;++ timeout = -1L;++ if (got_SIGHUP)+ {+ got_SIGHUP = false;+ ProcessConfigFile(PGC_SIGHUP);+ }++ if (got_SIGUSR2)+ /* It is needed only for wakeup worker */+ got_SIGUSR2 = false;++ /*+ * Move task items from shared buffer to local and release it for+ * new data.+ * It is introduced in accordance with the idea that shared memory+ * buffer is smaller than internal process buffer.+ */+ LWLockAcquire(&MyWorkerInfo->WorkItemLock, LW_SHARED);+ if (MyWorkerInfo->nitems > 0)+ {+ task_items_num = MyWorkerInfo->nitems;+ memcpy(task_item, MyWorkerInfo->buffer, task_items_num*sizeof(CleanerMessage));+ MyWorkerInfo->nitems = 0;+ }+ LWLockRelease(&MyWorkerInfo->WorkItemLock);++ /*+ * If launcher receive some task items it need to distribute between+ * workers and waiting lists.+ */+ if (task_items_num > 0)+ {+ DirtyRelation *hashent;+ bool found;+ int i;++ pgstat_progress_update_param(PROGRESS_CLEANER_MISSED_BLOCKS, stat_missed_blocks);++ /* Pass across items and sort by relation */+ for (i = 0; i < task_items_num; i++)+ {+ WorkerTask *item;++ hashent = SHASH_Search(PrivateRelationsTable,+ (void *) &(task_item[i].relid), HASH_FIND, NULL);++ if (hashent == NULL)+ {+ elog(LOG, "Add new relation: %d", task_item[i].relid);+ if (dirty_relations_num == WORKER_RELATIONS_MAX_NUM)+ {+ int j;+ int min = SHASH_Entries(dirty_relation[0]->items);+ int pos = 0;++ /*+ * All slots for dirty relations are busy.+ * Search for slot with smaller number of dirty blocks+ * and replace it.+ */+ elog(LOG, "All dirty relation slots are busy at worker process. Start min=%d", min);+ for (j = 1; j < dirty_relations_num; j++)+ {+ int nitems = SHASH_Entries(dirty_relation[j]->items);+ if (nitems < min)+ {+ min = nitems;+ pos = j;+ }+ }++ /* Save stats about not cleaned blocks */+ stat_missed_blocks += SHASH_Entries(dirty_relation[pos]->items);++ elog(LOG, "Remove slot %d with id=%d nitems=%lu. Set new relation id=%d",pos, dirty_relation[pos]->relid, SHASH_Entries(dirty_relation[pos]->items), task_item[i].relid);+ hashent = SHASH_Search(PrivateRelationsTable,+ (void *) &(dirty_relation[pos]->relid), HASH_REMOVE, &found);++ Assert(found);+ Assert(hashent != NULL);+ SHASH_Destroy(hashent->items);++ /*+ * Fill empty position by data from last position+ */+ dirty_relation[pos] = dirty_relation[--dirty_relations_num];+ }++ /* Insert new relid to hash table */+ hashent = SHASH_Search(PrivateRelationsTable, (void *) &(task_item[i].relid), HASH_ENTER, &found);+ Assert(!found);+ Assert(hashent != NULL);++ /*+ * At a new entry we create and init list of 'dirty' blocks+ */+ hashent->items = SHASH_Create(shctl);+ dirty_relation[dirty_relations_num++] = hashent;+ pgstat_progress_update_param(PROGRESS_CLEANER_RELATIONS, SHASH_Entries(PrivateRelationsTable));+ }++ /*+ * Relation entry found or create.+ * Now, add an item to a waiting list.+ */+ item = (WorkerTask *) SHASH_Search(hashent->items, (void *) &(task_item[i].blkno), HASH_ENTER, &found);+ if (item == NULL)+ stat_missed_blocks++;+ else+ {+ if (!found)+ {+ item->hits = 0;+ item->lastXid = InvalidTransactionId;+ }++ item->hits += task_item[i].hits;+ Assert(item->hits > 0);+ item->lastXid = (item->lastXid > task_item[i].xid) ? item->lastXid : task_item[i].xid;+ }+ }+ task_items_num = 0;+ }++ PG_TRY();+ {+ int relcounter;+ int64 stat_tot_wait_queue_len = 0;++ /* Pass along dirty relations and try to clean it */+ for (relcounter = 0; relcounter < dirty_relations_num; relcounter++)+ {+ if (SHASH_Entries(FreeDirtyBlocksList) > 0)+ SHASH_Clean(FreeDirtyBlocksList);++ FreeDirtyBlocksList = cleanup_relations(dirty_relation[relcounter], FreeDirtyBlocksList, got_SIGTERM);++ /*+ * Some blocks from the list may blocked by a backend.+ * Deferred its for next cleanup attempt.+ */+ if ((stat_tot_wait_queue_len += SHASH_Entries(dirty_relation[relcounter]->items)) > 0)+ timeout = 10L;+ }++ pgstat_progress_update_param(PROGRESS_CLEANER_TOTAL_QUEUE_LENGTH, stat_tot_wait_queue_len);+ QueryCancelPending = false;+ }+ PG_CATCH();+ {+ HOLD_INTERRUPTS();+ EmitErrorReport();+ AbortOutOfAnyTransaction();+ FlushErrorState();++ RESUME_INTERRUPTS();+ }+ PG_END_TRY();++ if (!got_SIGTERM)+ {+ int wakeEvents = WL_LATCH_SET | WL_POSTMASTER_DEATH;+ if (timeout > 0)+ wakeEvents |= WL_TIMEOUT;+ rc = WaitLatch(MyLatch, wakeEvents, timeout, WAIT_EVENT_BGHEAP_MAIN);+ }++ ResetLatch(MyLatch);++ /* Emergency bailout if postmaster has died */+ if (rc & WL_POSTMASTER_DEATH)+ proc_exit(1);+ }++ pgstat_progress_end_command();+ proc_exit(0);+}++/*+ * SIGHUP_Handler+ */+static void+SIGHUP_Handler(SIGNAL_ARGS)+{+ int save_errno = errno;++ got_SIGHUP = true;+ SetLatch(MyLatch);++ errno = save_errno;+}++/*+ * SIGTERM_Handler+ */+static void+SIGTERM_Handler(SIGNAL_ARGS)+{+ int save_errno = errno;++ got_SIGTERM = true;++ SetLatch(MyLatch);++ errno = save_errno;+}++/*+ * SIGUSR2_Handler+ * Wake up a worker to read some messages from launcher+ */+static void+SIGUSR2_Handler(SIGNAL_ARGS)+{+ int save_errno = errno;++ got_SIGUSR2 = true;++ SetLatch(MyLatch);++ errno = save_errno;+}++/*+ * Main entry point for background heap cleaner (launcher) process, to be called from the+ * postmaster.+ */+int+StartHeapCleanerLauncher(void)+{+ pid_t HeapCleanerLauncherPID;++#ifdef EXEC_BACKEND+ switch ((HeapCleanerLauncherPID = bghclauncher_forkexec()))+#else+ switch ((HeapCleanerLauncherPID = fork_process()))+#endif+ {+ case -1:+ ereport(LOG,+ (errmsg("could not fork background heap cleaner (launcher) process: %m")));+ return 0;++#ifndef EXEC_BACKEND+ case 0:+ /* in postmaster child ... */+ InitPostmasterChild();++ /* Close the postmaster's sockets */+ ClosePostmasterPorts(false);++ HeapCleanerLauncherMain(0, NULL);+ break;+#endif+ default:+ return (int) HeapCleanerLauncherPID;+ }++ /* shouldn't get here */+ return 0;+}++/*+ * Main entry point for background heap cleaner (worker) process, to be called from the+ * postmaster.+ */+int+StartHeapCleanerWorker(void)+{+ pid_t worker_pid;++#ifdef EXEC_BACKEND+ switch ((worker_pid = hcworker_forkexec()))+#else+ switch ((worker_pid = fork_process()))+#endif+ {+ case -1:+ ereport(LOG,+ (errmsg("could not fork background heap cleaner (worker) process: %m")));+ return 0;++#ifndef EXEC_BACKEND+ case 0:+ /* in postmaster child ... */+ InitPostmasterChild();++ /* Close the postmaster's sockets */+ ClosePostmasterPorts(false);++ HeapCleanerWorkerMain(0, NULL);+ break;+#endif+ default:+ return (int) worker_pid;+ }++ /* shouldn't get here */+ return 0;+}++/*+ * HeapCleanerShmemInit+ * Allocate and initialize heapcleaner-related shared memory+ */+void+HeapCleanerShmemInit(void)+{+ bool found;++ HeapCleanerShmem = (HeapCleanerShmemStruct *)+ ShmemInitStruct("HeapCleaner Data",+ HeapCleanerShmemSize(),+ &found);++ if (!IsUnderPostmaster)+ {+ WorkerInfo worker;+ int i;++ Assert(!found);++ dlist_init(&HeapCleanerShmem->freeWorkers);+ dlist_init(&HeapCleanerShmem->runningWorkers);++ worker = (WorkerInfo) ((char *) HeapCleanerShmem ++ MAXALIGN(sizeof(HeapCleanerShmemStruct)));++ /* initialize the WorkerInfo free list */+ for (i = 0; i < heapcleaner_max_workers; i++)+ {+ worker[i].id = i;+ LWLockInitialize(&worker[i].WorkItemLock, TrancheId);+ dlist_push_head(&HeapCleanerShmem->freeWorkers,+ &worker[i].links);+ }+ }+ else+ Assert(found);+ HeapCleanerShmem->startingWorker = NULL;++ TrancheId = LWLockNewTrancheId();+ LWLockRegisterTranche(TrancheId, "heapcleaner");+}++/*+ * HeapCleanerShmemSiz+ * Compute space needed for heap cleaner-related shared memory+ */+Size+HeapCleanerShmemSize(void)+{+ Size size;++ /*+ * Need the fixed struct and the array of WorkerInfoData.+ */+ size = sizeof(HeapCleanerShmemStruct);+ size = MAXALIGN(size);+ size = add_size(size, mul_size(heapcleaner_max_workers,+ sizeof(WorkerInfoData)));+ return size;+}++static uint64+CleanerMessageHashFunc(void *key, uint64 size, uint64 base)+{+ uint64 sum = 0;+ CleanerMessage *msg = (CleanerMessage *) key;++ Assert(key != NULL);+ Assert(size == 2 * sizeof(Oid) + sizeof(BlockNumber));+ Assert(base > 0);++ sum += DefaultHashValueFunc(&msg->dbNode, sizeof(Oid), base);+ sum += DefaultHashValueFunc(&msg->relid, sizeof(Oid), base);+ sum += DefaultHashValueFunc(&msg->blkno, sizeof(BlockNumber), base);++ return sum%base;+}diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.cindex bbe73618c7..f992307e37 100644--- a/src/backend/postmaster/pgstat.c+++ b/src/backend/postmaster/pgstat.c@@ -48,6 +48,7 @@
#include "miscadmin.h"
#include "pg_trace.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "postmaster/fork_process.h"
#include "postmaster/postmaster.h"
#include "replication/walsender.h"
@@ -1394,6 +1395,20 @@ pgstat_report_autovac(Oid dboid)
pgstat_send(&msg, sizeof(msg));
}
+void+pgstat_report_heapcleaner(Oid dboid)+{+ PgStat_MsgAutovacStart msg;++ if (pgStatSock == PGINVALID_SOCKET)+ return;++ pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_BGHEAP_START);+ msg.m_databaseid = dboid;+ msg.m_start_time = GetCurrentTimestamp();++ pgstat_send(&msg, sizeof(msg));+}
/* ---------
* pgstat_report_vacuum() -
@@ -2833,6 +2848,14 @@ pgstat_bestart(void)
/* Autovacuum Worker */
beentry->st_backendType = B_AUTOVAC_WORKER;
}
+ else if (IsHeapCleanerLauncherProcess())+ {+ beentry->st_backendType = B_BG_HEAPCLNR_LAUNCHER;+ }+ else if (IsHeapCleanerWorkerProcess())+ {+ beentry->st_backendType = B_BG_HEAPCLNR_WORKER;+ }
else if (am_walsender)
{
/* Wal sender */
@@ -3483,6 +3506,9 @@ pgstat_get_wait_activity(WaitEventActivity w)
case WAIT_EVENT_AUTOVACUUM_MAIN:
event_name = "AutoVacuumMain";
break;
+ case WAIT_EVENT_BGHEAP_MAIN:+ event_name = "BgHeapMain";+ break;
case WAIT_EVENT_BGWRITER_HIBERNATE:
event_name = "BgWriterHibernate";
break;
@@ -4111,6 +4137,12 @@ pgstat_get_backend_desc(BackendType backendType)
case B_AUTOVAC_WORKER:
backendDesc = "autovacuum worker";
break;
+ case B_BG_HEAPCLNR_LAUNCHER:+ backendDesc = "heap cleaner launcher";+ break;+ case B_BG_HEAPCLNR_WORKER:+ backendDesc = "heap cleaner worker";+ break;
case B_BACKEND:
backendDesc = "client backend";
break;
@@ -4422,6 +4454,10 @@ PgstatCollectorMain(int argc, char *argv[])
pgstat_recv_autovac((PgStat_MsgAutovacStart *) &msg, len);
break;
+ case PGSTAT_MTYPE_BGHEAP_START:+// pgstat_recv_autovac((PgStat_MsgHeapCleanerStart *) &msg, len);+ break;+
case PGSTAT_MTYPE_VACUUM:
pgstat_recv_vacuum((PgStat_MsgVacuum *) &msg, len);
break;
diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.cindex a4b53b33cd..d14c1b557d 100644--- a/src/backend/postmaster/postmaster.c+++ b/src/backend/postmaster/postmaster.c@@ -109,6 +109,7 @@
#include "pgstat.h"
#include "port/pg_bswap.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "postmaster/bgworker_internals.h"
#include "postmaster/fork_process.h"
#include "postmaster/pgarch.h"
@@ -145,9 +146,10 @@
#define BACKEND_TYPE_AUTOVAC 0x0002 /* autovacuum worker process */
#define BACKEND_TYPE_WALSND 0x0004 /* walsender process */
#define BACKEND_TYPE_BGWORKER 0x0008 /* bgworker process */
-#define BACKEND_TYPE_ALL 0x000F /* OR of all the above */+#define BACKEND_TYPE_BGHEAP 0x0010 /* autovacuum worker process */+#define BACKEND_TYPE_ALL 0x001F /* OR of all the above */-#define BACKEND_TYPE_WORKER (BACKEND_TYPE_AUTOVAC | BACKEND_TYPE_BGWORKER)+#define BACKEND_TYPE_WORKER (BACKEND_TYPE_AUTOVAC | BACKEND_TYPE_BGWORKER | BACKEND_TYPE_BGHEAP)
/*
* List of active backends (or child processes anyway; we don't actually
@@ -252,6 +254,7 @@ static pid_t StartupPID = 0,
WalWriterPID = 0,
WalReceiverPID = 0,
AutoVacPID = 0,
+ HeapClnrPID = 0,
PgArchPID = 0,
PgStatPID = 0,
SysLoggerPID = 0;
@@ -355,6 +358,7 @@ bool redirection_done = false; /* stderr redirected for syslogger? */
/* received START_AUTOVAC_LAUNCHER signal */
static volatile sig_atomic_t start_autovac_launcher = false;
+static volatile sig_atomic_t start_heapclnr_launcher = false;
/* the launcher needs to be signalled to communicate some condition */
static volatile bool avlauncher_needs_signal = false;
@@ -432,6 +436,7 @@ static void maybe_start_bgworkers(void);
static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
static pid_t StartChildProcess(AuxProcType type);
static void StartAutovacuumWorker(void);
+static void StartBgHeapWorker(void);
static void MaybeStartWalReceiver(void);
static void InitPostmasterDeathWatchHandle(void);
@@ -1310,6 +1315,7 @@ PostmasterMain(int argc, char *argv[])
*/
autovac_init();
+ HeapCleanerInit();
/*
* Load configuration files for client authentication.
*/
@@ -1757,6 +1763,14 @@ ServerLoop(void)
start_autovac_launcher = false; /* signal processed */
}
+ if (!IsBinaryUpgrade && HeapClnrPID == 0 && start_heapclnr_launcher &&+ pmState == PM_RUN)+ {+ HeapClnrPID = StartHeapCleanerLauncher();+ if (HeapClnrPID != 0)+ start_heapclnr_launcher = false; /* signal processed */+ }+
/* If we have lost the stats collector, try to start a new one */
if (PgStatPID == 0 &&
(pmState == PM_RUN || pmState == PM_HOT_STANDBY))
@@ -1774,6 +1788,12 @@ ServerLoop(void)
kill(AutoVacPID, SIGUSR2);
}
+ /* If we have lost the background heap cleaner, try to start a new one. */+ if (!IsBinaryUpgrade && HeapClnrPID == 0 && pmState == PM_RUN )+ {+ HeapClnrPID = StartHeapCleanerLauncher();+ }+
/* If we need to start a WAL receiver, try to do that now */
if (WalReceiverRequested)
MaybeStartWalReceiver();
@@ -2541,6 +2561,8 @@ SIGHUP_handler(SIGNAL_ARGS)
signal_child(WalReceiverPID, SIGHUP);
if (AutoVacPID != 0)
signal_child(AutoVacPID, SIGHUP);
+ if (HeapClnrPID != 0)+ signal_child(HeapClnrPID, SIGHUP);
if (PgArchPID != 0)
signal_child(PgArchPID, SIGHUP);
if (SysLoggerPID != 0)
@@ -2627,10 +2649,12 @@ pmdie(SIGNAL_ARGS)
/* autovac workers are told to shut down immediately */
/* and bgworkers too; does this need tweaking? */
SignalSomeChildren(SIGTERM,
- BACKEND_TYPE_AUTOVAC | BACKEND_TYPE_BGWORKER);+ BACKEND_TYPE_AUTOVAC | BACKEND_TYPE_BGWORKER | BACKEND_TYPE_BGHEAP);
/* and the autovac launcher too */
if (AutoVacPID != 0)
signal_child(AutoVacPID, SIGTERM);
+ if (HeapClnrPID != 0)+ signal_child(HeapClnrPID, SIGTERM);
/* and the bgwriter too */
if (BgWriterPID != 0)
signal_child(BgWriterPID, SIGTERM);
@@ -2687,7 +2711,7 @@ pmdie(SIGNAL_ARGS)
signal_child(WalReceiverPID, SIGTERM);
if (pmState == PM_RECOVERY)
{
- SignalSomeChildren(SIGTERM, BACKEND_TYPE_BGWORKER);+ SignalSomeChildren(SIGTERM, BACKEND_TYPE_BGWORKER | BACKEND_TYPE_BGHEAP);
/*
* Only startup, bgwriter, walreceiver, possibly bgworkers,
@@ -2708,10 +2732,12 @@ pmdie(SIGNAL_ARGS)
/* shut down all backends and workers */
SignalSomeChildren(SIGTERM,
BACKEND_TYPE_NORMAL | BACKEND_TYPE_AUTOVAC |
- BACKEND_TYPE_BGWORKER);+ BACKEND_TYPE_BGWORKER | BACKEND_TYPE_BGHEAP);
/* and the autovac launcher too */
if (AutoVacPID != 0)
signal_child(AutoVacPID, SIGTERM);
+ if (HeapClnrPID != 0)+ signal_child(HeapClnrPID, SIGTERM);
/* and the walwriter too */
if (WalWriterPID != 0)
signal_child(WalWriterPID, SIGTERM);
@@ -2875,6 +2901,8 @@ reaper(SIGNAL_ARGS)
*/
if (!IsBinaryUpgrade && AutoVacuumingActive() && AutoVacPID == 0)
AutoVacPID = StartAutoVacLauncher();
+ if (!IsBinaryUpgrade && (HeapClnrPID == 0))+ HeapClnrPID = StartHeapCleanerLauncher();
if (PgArchStartupAllowed() && PgArchPID == 0)
PgArchPID = pgarch_start();
if (PgStatPID == 0)
@@ -3009,6 +3037,14 @@ reaper(SIGNAL_ARGS)
_("autovacuum launcher process"));
continue;
}
+ if (pid == HeapClnrPID)+ {+ HeapClnrPID = 0;+ if (!EXIT_STATUS_0(exitstatus))+ HandleChildCrash(pid, exitstatus,+ _("heap cleaner launcher process"));+ continue;+ }
/*
* Was it the archiver? If so, just try to start a new one; no need
@@ -3475,6 +3511,17 @@ HandleChildCrash(int pid, int exitstatus, const char *procname)
signal_child(AutoVacPID, (SendStop ? SIGSTOP : SIGQUIT));
}
+ if (pid == HeapClnrPID)+ HeapClnrPID = 0;+ else if (HeapClnrPID != 0 && take_action)+ {+ ereport(DEBUG2,+ (errmsg_internal("sending %s to process %d",+ "SIGQUIT",+ (int) HeapClnrPID)));+ signal_child(HeapClnrPID, SIGQUIT);+ }+
/*
* Force a power-cycle of the pgarch process too. (This isn't absolutely
* necessary, but it seems like a good idea for robustness, and it
@@ -3661,6 +3708,7 @@ PostmasterStateMachine(void)
(CheckpointerPID == 0 ||
(!FatalError && Shutdown < ImmediateShutdown)) &&
WalWriterPID == 0 &&
+ HeapClnrPID == 0 &&
AutoVacPID == 0)
{
if (Shutdown >= ImmediateShutdown || FatalError)
@@ -3759,6 +3807,7 @@ PostmasterStateMachine(void)
Assert(CheckpointerPID == 0);
Assert(WalWriterPID == 0);
Assert(AutoVacPID == 0);
+ Assert(HeapClnrPID == 0);
/* syslogger is not considered here */
pmState = PM_NO_CHILDREN;
}
@@ -3948,6 +3997,8 @@ TerminateChildren(int signal)
signal_child(WalReceiverPID, signal);
if (AutoVacPID != 0)
signal_child(AutoVacPID, signal);
+ if (HeapClnrPID != 0)+ signal_child(HeapClnrPID, signal);
if (PgArchPID != 0)
signal_child(PgArchPID, signal);
if (PgStatPID != 0)
@@ -4781,6 +4832,7 @@ SubPostmasterMain(int argc, char *argv[])
if (strcmp(argv[1], "--forkbackend") == 0 ||
strcmp(argv[1], "--forkavlauncher") == 0 ||
strcmp(argv[1], "--forkavworker") == 0 ||
+ strcmp(argv[1], "--forkbgheap") == 0 ||
strcmp(argv[1], "--forkboot") == 0 ||
strncmp(argv[1], "--forkbgworker=", 15) == 0)
PGSharedMemoryReAttach();
@@ -4792,6 +4844,8 @@ SubPostmasterMain(int argc, char *argv[])
AutovacuumLauncherIAm();
if (strcmp(argv[1], "--forkavworker") == 0)
AutovacuumWorkerIAm();
+ if (strcmp(argv[1], "--forkbgheap") == 0)+ BgHeapCleanerIAm();
/*
* Start our win32 signal implementation. This has to be done after we
@@ -4921,6 +4975,19 @@ SubPostmasterMain(int argc, char *argv[])
AutoVacWorkerMain(argc - 2, argv + 2); /* does not return */
}
+ if (strcmp(argv[1], "--forkbgheap") == 0)+ {+ /* Restore basic shared memory pointers */+ InitShmemAccess(UsedShmemSegAddr);++ /* Need a PGPROC to run CreateSharedMemoryAndSemaphores */+ InitProcess();++ /* Attach process to shared data structures */+ CreateSharedMemoryAndSemaphores(false, 0);++ BgHeapMain(argc, argv); /* does not return */+ }
if (strncmp(argv[1], "--forkbgworker=", 15) == 0)
{
int shmem_slot;
@@ -5122,6 +5189,12 @@ sigusr1_handler(SIGNAL_ARGS)
start_autovac_launcher = true;
}
+ if (CheckPostmasterSignal(PMSIGNAL_START_HEAPCLNR_LAUNCHER) &&+ Shutdown == NoShutdown)+ {+ start_heapclnr_launcher = true;+ }+
if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER) &&
Shutdown == NoShutdown)
{
@@ -5129,6 +5202,13 @@ sigusr1_handler(SIGNAL_ARGS)
StartAutovacuumWorker();
}
+ if (CheckPostmasterSignal(PMSIGNAL_START_HEAPCLNR_WORKER) &&+ Shutdown == NoShutdown)+ {+ /* The autovacuum launcher wants us to start a worker process. */+ StartBgHeapWorker();+ }+
if (CheckPostmasterSignal(PMSIGNAL_START_WALRECEIVER))
{
/* Startup Process wants us to start the walreceiver process. */
@@ -5470,6 +5550,73 @@ StartAutovacuumWorker(void)
}
}
+static void+StartBgHeapWorker(void)+{+ Backend *bn;++ /*+ * If not in condition to run a process, don't try, but handle it like a+ * fork failure. This does not normally happen, since the signal is only+ * supposed to be sent by autovacuum launcher when it's OK to do it, but+ * we have to check to avoid race-condition problems during DB state+ * changes.+ */+ if (canAcceptConnections() == CAC_OK)+ {+ /*+ * Compute the cancel key that will be assigned to this session. We+ * probably don't need cancel keys for autovac workers, but we'd+ * better have something random in the field to prevent unfriendly+ * people from sending cancels to them.+ */+ if (!RandomCancelKey(&MyCancelKey))+ {+ ereport(LOG,+ (errcode(ERRCODE_INTERNAL_ERROR),+ errmsg("could not generate random cancel key")));+ return;+ }++ bn = (Backend *) malloc(sizeof(Backend));+ if (bn)+ {+ bn->cancel_key = MyCancelKey;++ /* Autovac workers are not dead_end and need a child slot */+ bn->dead_end = false;+ bn->child_slot = MyPMChildSlot = AssignPostmasterChildSlot();+ bn->bgworker_notify = false;++ bn->pid = StartHeapCleanerWorker();++ if (bn->pid > 0)+ {+ bn->bkend_type = BACKEND_TYPE_BGHEAP;+ dlist_push_head(&BackendList, &bn->elem);+#ifdef EXEC_BACKEND+ ShmemBackendArrayAdd(bn);+#endif+ /* all OK */+ return;+ }++ /*+ * fork failed, fall through to report -- actual error message was+ * logged by StartAutoVacWorker+ */+ (void) ReleasePostmasterChildSlot(bn->child_slot);+ free(bn);+ }+ else+ ereport(LOG,+ (errcode(ERRCODE_OUT_OF_MEMORY),+ errmsg("out of memory")));+ }++ elog(ERROR, "Some BGHEAP Worker start error!");+}+
/*
* MaybeStartWalReceiver
* Start the WAL receiver process, if not running and our state allows.
@@ -5534,7 +5681,8 @@ CreateOptsFile(int argc, char *argv[], char *fullprogname)
int
MaxLivePostmasterChildren(void)
{
- return 2 * (MaxConnections + autovacuum_max_workers + 1 ++ return 2 * (MaxConnections + autovacuum_max_workers ++ heapcleaner_max_workers + 1 +
max_worker_processes);
}
diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.cindex 01eabe5706..175902e0d3 100644--- a/src/backend/storage/buffer/bufmgr.c+++ b/src/backend/storage/buffer/bufmgr.c@@ -449,6 +449,7 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
ForkNumber forkNum,
BlockNumber blockNum,
BufferAccessStrategy strategy,
+ bool onlyInMemory,
bool *foundPtr);
static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
static void AtProcExit_Buffers(int code, Datum arg);
@@ -717,6 +718,8 @@ ReadBuffer_common(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
isExtend = (blockNum == P_NEW);
+ Assert(!(isExtend && mode == RBM_NORMAL_NO_READ));+
TRACE_POSTGRESQL_BUFFER_READ_START(forkNum, blockNum,
smgr->smgr_rnode.node.spcNode,
smgr->smgr_rnode.node.dbNode,
@@ -743,9 +746,11 @@ ReadBuffer_common(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
* not currently in memory.
*/
bufHdr = BufferAlloc(smgr, relpersistence, forkNum, blockNum,
- strategy, &found);+ strategy, (mode == RBM_NORMAL_NO_READ), &found);
if (found)
pgBufferUsage.shared_blks_hit++;
+ else if (mode == RBM_NORMAL_NO_READ)+ return InvalidBuffer;
else
pgBufferUsage.shared_blks_read++;
}
@@ -788,6 +793,8 @@ ReadBuffer_common(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
return BufferDescriptorGetBuffer(bufHdr);
}
+ Assert(mode != RBM_NORMAL_NO_READ);+
/*
* We get here only in the corner case where we are trying to extend
* the relation but we found a pre-existing buffer marked BM_VALID.
@@ -987,6 +994,7 @@ static BufferDesc *
BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
BlockNumber blockNum,
BufferAccessStrategy strategy,
+ bool onlyInMemory,
bool *foundPtr)
{
BufferTag newTag; /* identity of requested block */
@@ -1036,7 +1044,17 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
* own read attempt if the page is still not BM_VALID.
* StartBufferIO does it all.
*/
- if (StartBufferIO(buf, true))+ if (onlyInMemory == true)+ {+ /*+ * We cannot get buffer immediately, go away if caller just+ * tried to get access to already in-memory page+ */+ UnpinBuffer(buf, true);+ *foundPtr = false;+ return NULL;+ }+ else if (StartBufferIO(buf, true))
{
/*
* If we get here, previous attempts to read the buffer must
@@ -1055,6 +1073,13 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
*/
LWLockRelease(newPartitionLock);
+ /* will not read a page from disk */+ if (onlyInMemory == true)+ {+ *foundPtr = false;+ return NULL;+ }+
/* Loop here in case we have to try another victim buffer */
for (;;)
{
@@ -1496,6 +1521,37 @@ MarkBufferDirty(Buffer buffer)
}
}
+/*+ * IsBufferDirty - is buffer marked dirty?+ *+ * Buffer should be pinned, if caller wishes to get consistent result,+ * buffer should be exclusively locked.+ */+bool+IsBufferDirty(Buffer buffer)+{+ BufferDesc *bufHdr;+ uint32 buf_state;++ if (!BufferIsValid(buffer))+ elog(ERROR, "bad buffer ID: %d", buffer);++ Assert(BufferIsPinned(buffer));++ if (BufferIsLocal(buffer))+ return IsLocalBufferDirty(buffer);++ bufHdr = GetBufferDescriptor(buffer - 1);++ buf_state = pg_atomic_read_u32(&bufHdr->state);++ /*+ * if buffer is not exclusively locked, buf_state could be out of sync+ * already...+ */+ return (buf_state & BM_DIRTY) ? true : false;+}+
/*
* ReleaseAndReadBuffer -- combine ReleaseBuffer() and ReadBuffer()
*
diff --git a/src/backend/storage/buffer/localbuf.c b/src/backend/storage/buffer/localbuf.cindex e4146a260a..856955635e 100644--- a/src/backend/storage/buffer/localbuf.c+++ b/src/backend/storage/buffer/localbuf.c@@ -305,6 +305,30 @@ MarkLocalBufferDirty(Buffer buffer)
pg_atomic_unlocked_write_u32(&bufHdr->state, buf_state);
}
+/*+ * IsLocalBufferDirty+ * returns true of buffer is dirty+ */+bool+IsLocalBufferDirty(Buffer buffer)+{+ int bufid;+ BufferDesc *bufHdr;+ uint32 buf_state;++ Assert(BufferIsLocal(buffer));++ bufid = -(buffer + 1);++ Assert(LocalRefCount[bufid] > 0);++ bufHdr = GetLocalBufferDescriptor(bufid);++ buf_state = pg_atomic_read_u32(&bufHdr->state);++ return (buf_state & BM_DIRTY) != 0;+}+
/*
* DropRelFileNodeLocalBuffers
* This function removes from the buffer pool all the pages of the
diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.cindex 0c86a581c0..fba74752a9 100644--- a/src/backend/storage/ipc/ipci.c+++ b/src/backend/storage/ipc/ipci.c@@ -25,6 +25,7 @@
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "postmaster/bgworker_internals.h"
#include "postmaster/bgwriter.h"
#include "postmaster/postmaster.h"
@@ -140,6 +141,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
size = add_size(size, ProcSignalShmemSize());
size = add_size(size, CheckpointerShmemSize());
size = add_size(size, AutoVacuumShmemSize());
+ size = add_size(size, HeapCleanerShmemSize());
size = add_size(size, ReplicationSlotsShmemSize());
size = add_size(size, ReplicationOriginShmemSize());
size = add_size(size, WalSndShmemSize());
@@ -256,6 +258,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
ProcSignalShmemInit();
CheckpointerShmemInit();
AutoVacuumShmemInit();
+ HeapCleanerShmemInit();
ReplicationSlotsShmemInit();
ReplicationOriginShmemInit();
WalSndShmemInit();
diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txtindex e6025ecedb..73a9d141b0 100644--- a/src/backend/storage/lmgr/lwlocknames.txt+++ b/src/backend/storage/lmgr/lwlocknames.txt@@ -50,3 +50,4 @@ OldSnapshotTimeMapLock 42
BackendRandomLock 43
LogicalRepWorkerLock 44
CLogTruncationLock 45
+HeapCleanerLock 46diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.cindex 6f30e082b2..ecf14647b4 100644--- a/src/backend/storage/lmgr/proc.c+++ b/src/backend/storage/lmgr/proc.c@@ -41,6 +41,7 @@
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "replication/slot.h"
#include "replication/syncrep.h"
#include "storage/condition_variable.h"
@@ -352,7 +353,7 @@ InitProcess(void)
* cleaning up. (XXX autovac launcher currently doesn't participate in
* this; it probably should.)
*/
- if (IsUnderPostmaster && !IsAutoVacuumLauncherProcess())+ if (IsUnderPostmaster && !IsAutoVacuumLauncherProcess() && !IsHeapCleanerLauncherProcess())
MarkPostmasterChildActive();
/*
@@ -896,7 +897,7 @@ ProcKill(int code, Datum arg)
* way, so tell the postmaster we've cleaned up acceptably well. (XXX
* autovac launcher should be included here someday)
*/
- if (IsUnderPostmaster && !IsAutoVacuumLauncherProcess())+ if (IsUnderPostmaster && !IsAutoVacuumLauncherProcess() && !IsHeapCleanerLauncherProcess())
MarkPostmasterChildInactive();
/* wake autovac launcher if needed -- see comments in FreeWorkerInfo */
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.cindex f4133953be..87779bd178 100644--- a/src/backend/tcop/postgres.c+++ b/src/backend/tcop/postgres.c@@ -56,6 +56,7 @@
#include "parser/parser.h"
#include "pg_getopt.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "postmaster/postmaster.h"
#include "replication/logicallauncher.h"
#include "replication/logicalworker.h"
@@ -2889,6 +2890,10 @@ ProcessInterrupts(void)
ereport(FATAL,
(errcode(ERRCODE_ADMIN_SHUTDOWN),
errmsg("terminating autovacuum process due to administrator command")));
+ else if (IsHeapCleanerWorkerProcess())+ ereport(FATAL,+ (errcode(ERRCODE_ADMIN_SHUTDOWN),+ errmsg("terminating bgheap process due to administrator command")));
else if (IsLogicalWorker())
ereport(FATAL,
(errcode(ERRCODE_ADMIN_SHUTDOWN),
@@ -3017,6 +3022,13 @@ ProcessInterrupts(void)
(errcode(ERRCODE_QUERY_CANCELED),
errmsg("canceling autovacuum task")));
}
+ if (IsHeapCleanerWorkerProcess())+ {+ LockErrorCleanup();+ ereport(ERROR,+ (errcode(ERRCODE_QUERY_CANCELED),+ errmsg("canceling bgheap worker task")));+ }
if (RecoveryConflictPending)
{
RecoveryConflictPending = false;
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.cindex e95e347184..8831d98460 100644--- a/src/backend/utils/adt/pgstatfuncs.c+++ b/src/backend/utils/adt/pgstatfuncs.c@@ -468,6 +468,8 @@ pg_stat_get_progress_info(PG_FUNCTION_ARGS)
/* Translate command name into command type code. */
if (pg_strcasecmp(cmd, "VACUUM") == 0)
cmdtype = PROGRESS_COMMAND_VACUUM;
+ else if (pg_strcasecmp(cmd, "CLEANER") == 0)+ cmdtype = PROGRESS_COMMAND_CLEANER;
else
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
diff --git a/src/backend/utils/hash/Makefile b/src/backend/utils/hash/Makefileindex 64eebd1d99..9a3392fe37 100644--- a/src/backend/utils/hash/Makefile+++ b/src/backend/utils/hash/Makefile@@ -12,6 +12,6 @@ subdir = src/backend/utils/hash
top_builddir = ../../../..
include $(top_builddir)/src/Makefile.global
-OBJS = dynahash.o hashfn.o pg_crc.o+OBJS = dynahash.o hashfn.o pg_crc.o shash.o
include $(top_srcdir)/src/backend/common.mk
diff --git a/src/backend/utils/hash/shash.c b/src/backend/utils/hash/shash.cnew file mode 100644index 0000000000..9710485c85--- /dev/null+++ b/src/backend/utils/hash/shash.c@@ -0,0 +1,265 @@+/*+ * shash.c+ *+ * Lightweight hash tables without dynamic expansion+ *+ *+ * Copyright (c) 1996-2018, PostgreSQL Global Development Group+ *+ *+ * IDENTIFICATION+ * src/backend/utils/hash/shash.c+ *+ *-------------------------------------------------------------------------+ */++#include "postgres.h"++#include "limits.h"+#include "utils/shash.h"++#define ELEM(index) (&shtab->Elements[index*shtab->Header.ElementSize])++static void check(SHTAB* shtab);++/*+ * Check data allocation of hash table variables+ */+static void+check(SHTAB* shtab)+{+ Assert(shtab != NULL);+ Assert(shtab->Elements != NULL);+ Assert(shtab->state != NULL);+}++/*+ * Create and initialize hash table with shctl options+ */+SHTAB*+SHASH_Create(SHTABCTL shctl)+{+ SHTAB* shtab = (SHTAB *) palloc(sizeof(SHTAB));++ Assert(shctl.ElementSize != 0);+ Assert(shctl.ElementSize >= shctl.KeySize);++ shtab->Header = shctl;+ shtab->HTableSize = shtab->Header.ElementsMaxNum * (2. - shtab->Header.FillFactor) + 1;+ Assert(shtab->HTableSize > shtab->Header.ElementsMaxNum);++ /* Add one element as sign of empty value */+ shtab->Elements = (char *) palloc(shtab->HTableSize * shctl.ElementSize);+ shtab->state = (HESTATE *) palloc0(shtab->HTableSize * sizeof(HESTATE));+ shtab->nElements = 0;+ shtab->SeqScanCurElem = 0;++ return shtab;+}++/*+ * Fast cleaning of hash table.+ * After call of this function hash table contains 0 elements+ * and can be used again.+ */+void+SHASH_Clean(SHTAB* shtab)+{+ int i;++ check(shtab);++ /*+ * We do not free memory: only clean 'used' field and set number of elems at 0+ */+ for (i = 0; i < shtab->HTableSize; i++)+ shtab->state[i] = SHASH_NUSED;+ shtab->nElements = 0;+ shtab->SeqScanCurElem = 0;+}++/*+ * Free all hash table memory resources.+ */+void+SHASH_Destroy(SHTAB* shtab)+{+ check(shtab);++ pfree(shtab->Elements);+ pfree(shtab->state);+ pfree(shtab);+}++/*+ * Get number of elements in the hash table.+ */+uint64+SHASH_Entries(SHTAB* shtab)+{+ check(shtab);++ Assert((shtab->nElements >= 0) && (shtab->nElements <= shtab->Header.ElementsMaxNum));+ return shtab->nElements;+}++/*+ * Find, enter or delete element.+ * return NULL, if element not found.+ */+void*+SHASH_Search(SHTAB* shtab, void *keyPtr, SHASHACTION action, bool *foundPtr)+{+ uint64 index,+ first,+ first_removed_index = ULONG_MAX;+ void *result = NULL;++ check(shtab);++ Assert(shtab->nElements <= shtab->Header.ElementsMaxNum);++ first = index = shtab->Header.HashFunc(keyPtr, shtab->Header.KeySize, shtab->HTableSize);+ Assert(index < shtab->HTableSize);++ if (foundPtr != NULL)+ *foundPtr = false;++ /*+ * Main hash table search cycle+ */+ for(;;)+ {+ if (shtab->state[index] == SHASH_NUSED)+ {+ /* Empty position found */+ switch (action)+ {+ case SHASH_ENTER:+ if (shtab->nElements < shtab->Header.ElementsMaxNum)+ {+ if (first_removed_index != ULONG_MAX)+ index= first_removed_index;++ memcpy(ELEM(index), keyPtr, shtab->Header.KeySize);+ shtab->state[index] = SHASH_USED;+ shtab->nElements++;+ result = ELEM(index);+ }+ break;+ case SHASH_FIND:+ /* Hash table not contains record with such key */+ case SHASH_REMOVE:+ /* may be potential problem */+ break;+ default:+ Assert(0);+ }++ return result;+ }++ if ((shtab->state[index] == SHASH_USED) && (shtab->Header.CompFunc(keyPtr, ELEM(index))))+ {+ /*+ * We found the element exactly+ */+ if (foundPtr != NULL)+ *foundPtr = true;++ if (action == SHASH_REMOVE)+ {+ Assert(shtab->nElements > 0);+ /* Data will be cleaned by ENTER operation */+ shtab->state[index] = SHASH_REMOVED;+ shtab->nElements--;+ }++ return ELEM(index);+ }++ if ((shtab->state[index] == SHASH_REMOVED) && (action == SHASH_ENTER))+ /* We can use this element for SHASH_ENTER, potentially */+ if (first_removed_index == ULONG_MAX)+ first_removed_index = index;++ /* Go to next element */+ index = (index+1)%(shtab->HTableSize);++ if (index == first)+ {+ /*+ * We made all-table search cycle.+ */+ if ((action == SHASH_ENTER) && (first_removed_index != ULONG_MAX) && (shtab->nElements < shtab->Header.ElementsMaxNum))+ {+ index = first_removed_index;+ memcpy(ELEM(index), keyPtr, shtab->Header.KeySize);+ shtab->state[index] = SHASH_USED;+ shtab->nElements++;+ result = ELEM(index);+ }+ else+ result = NULL;+ return result;+ }+ }++ Assert(0);+ return NULL;+}++/*+ * Begin new sequental scan of the hash table.+ */+void+SHASH_SeqReset(SHTAB* shtab)+{+ check(shtab);++ shtab->SeqScanCurElem = 0;+}++/*+ * Get next hash table value during sequental scan.+ * Call after last element returned returns NULL.+ */+void*+SHASH_SeqNext(SHTAB* shtab)+{+ check(shtab);++ if (shtab->SeqScanCurElem == shtab->HTableSize)+ return NULL;++ Assert(shtab->SeqScanCurElem < shtab->HTableSize);++ do+ {+ if (shtab->state[shtab->SeqScanCurElem] == SHASH_USED)+ return ELEM(shtab->SeqScanCurElem++);+ else+ shtab->SeqScanCurElem++;+ }+ while (shtab->SeqScanCurElem < shtab->HTableSize);++ return NULL;+}++/*+ * Simple hash function.+ */+uint64+DefaultHashValueFunc(void *key, uint64 size, uint64 base)+{+ uint64 sum = 0;+ int i;++ for (i = 0; i < size; i++)+ {+ uint64 x = ((char *)key)[i];+ sum += x*x*x*x + x*x*x + x*x + x + 1;+ }++ return sum%base;+}diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.cindex 865119d272..9ebdd1c7de 100644--- a/src/backend/utils/init/miscinit.c+++ b/src/backend/utils/init/miscinit.c@@ -38,6 +38,7 @@
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "postmaster/postmaster.h"
#include "storage/fd.h"
#include "storage/ipc.h"
@@ -689,7 +690,7 @@ InitializeSessionUserIdStandalone(void)
* This function should only be called in single-user mode, in autovacuum
* workers, and in background workers.
*/
- AssertState(!IsUnderPostmaster || IsAutoVacuumWorkerProcess() || IsBackgroundWorker);+ AssertState(!IsUnderPostmaster || IsAutoVacuumWorkerProcess() || IsBackgroundWorker || IsHeapCleanerWorkerProcess());
/* call only once */
AssertState(!OidIsValid(AuthenticatedUserId));
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.cindex 5ef6315d20..4345b8bbbc 100644--- a/src/backend/utils/init/postinit.c+++ b/src/backend/utils/init/postinit.c@@ -38,6 +38,7 @@
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/autovacuum.h"
+#include "postmaster/bgheap.h"
#include "postmaster/postmaster.h"
#include "replication/walsender.h"
#include "storage/bufmgr.h"
@@ -321,7 +322,7 @@ CheckMyDatabase(const char *name, bool am_superuser, bool override_allow_connect
*
* We do not enforce them for autovacuum worker processes either.
*/
- if (IsUnderPostmaster && !IsAutoVacuumWorkerProcess())+ if (IsUnderPostmaster && !IsAutoVacuumWorkerProcess() && !IsHeapCleanerWorkerProcess())
{
/*
* Check that the database is currently allowing connections.
@@ -504,8 +505,8 @@ InitializeMaxBackends(void)
Assert(MaxBackends == 0);
/* the extra unit accounts for the autovacuum launcher */
- MaxBackends = MaxConnections + autovacuum_max_workers + 1 +- max_worker_processes;+ MaxBackends = MaxConnections + autovacuum_max_workers ++ heapcleaner_max_workers + 1 + max_worker_processes;
/* internal error because the values were all checked previously */
if (MaxBackends > MAX_BACKENDS)
@@ -682,7 +683,7 @@ InitPostgres(const char *in_dbname, Oid dboid, const char *username,
before_shmem_exit(ShutdownPostgres, 0);
/* The autovacuum launcher is done here */
- if (IsAutoVacuumLauncherProcess())+ if (IsAutoVacuumLauncherProcess() || IsHeapCleanerLauncherProcess())
{
/* report this backend in the PgBackendStatus array */
pgstat_bestart();
@@ -722,7 +723,7 @@ InitPostgres(const char *in_dbname, Oid dboid, const char *username,
* In standalone mode and in autovacuum worker processes, we use a fixed
* ID, otherwise we figure it out from the authenticated user name.
*/
- if (bootstrap || IsAutoVacuumWorkerProcess())+ if (bootstrap || IsAutoVacuumWorkerProcess() || IsHeapCleanerWorkerProcess())
{
InitializeSessionUserIdStandalone();
am_superuser = true;
diff --git a/src/include/commands/progress.h b/src/include/commands/progress.hindex 6a6b467fee..0cd51ad315 100644--- a/src/include/commands/progress.h+++ b/src/include/commands/progress.h@@ -34,4 +34,12 @@
#define PROGRESS_VACUUM_PHASE_TRUNCATE 5
#define PROGRESS_VACUUM_PHASE_FINAL_CLEANUP 6
+/* Progress parameters for cleaner */+#define PROGRESS_CLEANER_RELATIONS 0+#define PROGRESS_CLEANER_TOTAL_QUEUE_LENGTH 1+#define PROGRESS_CLEANER_FALSE_HITS 2+#define PROGRESS_CLEANER_TOTAL_DELETIONS 3+#define PROGRESS_CLEANER_VAINLY_CLEANED_TUPLES 4+#define PROGRESS_CLEANER_MISSED_BLOCKS 5+
#endif
diff --git a/src/include/commands/vacuum.h b/src/include/commands/vacuum.hindex 85d472f0a5..705120eddc 100644--- a/src/include/commands/vacuum.h+++ b/src/include/commands/vacuum.h@@ -186,6 +186,9 @@ extern void vacuum_set_xid_limits(Relation rel,
extern void vac_update_datfrozenxid(void);
extern void vacuum_delay_point(void);
+extern void quick_vacuum_index(Relation irel, Relation hrel,+ ItemPointer dead_tuples,+ int num_dead_tuples);
/* in commands/vacuumlazy.c */
extern void lazy_vacuum_rel(Relation onerel, int options,
VacuumParams *params, BufferAccessStrategy bstrategy);
diff --git a/src/include/pgstat.h b/src/include/pgstat.hindex d59c24ae23..dbf09328ac 100644--- a/src/include/pgstat.h+++ b/src/include/pgstat.h@@ -56,6 +56,7 @@ typedef enum StatMsgType
PGSTAT_MTYPE_RESETSHAREDCOUNTER,
PGSTAT_MTYPE_RESETSINGLECOUNTER,
PGSTAT_MTYPE_AUTOVAC_START,
+ PGSTAT_MTYPE_BGHEAP_START,
PGSTAT_MTYPE_VACUUM,
PGSTAT_MTYPE_ANALYZE,
PGSTAT_MTYPE_ARCHIVER,
@@ -354,6 +355,12 @@ typedef struct PgStat_MsgAutovacStart
TimestampTz m_start_time;
} PgStat_MsgAutovacStart;
+typedef struct PgStat_MsgHeapCleanerStart+{+ PgStat_MsgHdr m_hdr;+ Oid m_databaseid;+ TimestampTz m_start_time;+} PgStat_MsgHeapCleanerStart;
/* ----------
* PgStat_MsgVacuum Sent by the backend or autovacuum daemon
@@ -704,6 +711,8 @@ typedef enum BackendType
B_AUTOVAC_LAUNCHER,
B_AUTOVAC_WORKER,
B_BACKEND,
+ B_BG_HEAPCLNR_LAUNCHER,+ B_BG_HEAPCLNR_WORKER,
B_BG_WORKER,
B_BG_WRITER,
B_CHECKPOINTER,
@@ -756,6 +765,7 @@ typedef enum
{
WAIT_EVENT_ARCHIVER_MAIN = PG_WAIT_ACTIVITY,
WAIT_EVENT_AUTOVACUUM_MAIN,
+ WAIT_EVENT_BGHEAP_MAIN,
WAIT_EVENT_BGWRITER_HIBERNATE,
WAIT_EVENT_BGWRITER_MAIN,
WAIT_EVENT_CHECKPOINTER_MAIN,
@@ -933,7 +943,8 @@ typedef enum
typedef enum ProgressCommandType
{
PROGRESS_COMMAND_INVALID,
- PROGRESS_COMMAND_VACUUM+ PROGRESS_COMMAND_VACUUM,+ PROGRESS_COMMAND_CLEANER
} ProgressCommandType;
#define PGSTAT_NUM_PROGRESS_PARAM 10
@@ -1181,6 +1192,7 @@ extern void pgstat_reset_shared_counters(const char *);
extern void pgstat_reset_single_counter(Oid objectid, PgStat_Single_Reset_Type type);
extern void pgstat_report_autovac(Oid dboid);
+extern void pgstat_report_heapcleaner(Oid dboid);
extern void pgstat_report_vacuum(Oid tableoid, bool shared,
PgStat_Counter livetuples, PgStat_Counter deadtuples);
extern void pgstat_report_analyze(Relation rel,
diff --git a/src/include/postmaster/bgheap.h b/src/include/postmaster/bgheap.hnew file mode 100644index 0000000000..e11b310ffd--- /dev/null+++ b/src/include/postmaster/bgheap.h@@ -0,0 +1,36 @@+/*+ * bgheap.h+ *+ * Created on: 27.06.2018+ * Author: andrey+ */++#ifndef SRC_INCLUDE_ACCESS_BGHEAP_H_+#define SRC_INCLUDE_ACCESS_BGHEAP_H_++#include "storage/block.h"++#ifdef EXEC_BACKEND+extern void HeapCleanerLauncherIAm(void);+extern void HeapCleanerWorkerIAm(void);+extern void HeapCleanerLauncherMain(int argc, char *argv[]) pg_attribute_noreturn();+extern void HeapCleanerWorkerMain(int argc, char *argv[]) pg_attribute_noreturn();+#endif++extern int heapcleaner_max_workers;+++extern void AtEOXact_BGHeap_tables(bool isCommit);+extern bool IsHeapCleanerLauncherProcess(void);+extern bool IsHeapCleanerWorkerProcess(void);++extern void HeapCleanerInit(void);+extern int StartHeapCleanerLauncher(void);+extern int StartHeapCleanerWorker(void);++extern void HeapCleanerSend(Relation relation, BlockNumber blkno);++extern void HeapCleanerShmemInit(void);+extern Size HeapCleanerShmemSize(void);++#endif /* SRC_INCLUDE_ACCESS_BGHEAP_H_ */diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.hindex 5370035f0c..58be48198b 100644--- a/src/include/storage/buf_internals.h+++ b/src/include/storage/buf_internals.h@@ -333,6 +333,7 @@ extern void LocalPrefetchBuffer(SMgrRelation smgr, ForkNumber forkNum,
extern BufferDesc *LocalBufferAlloc(SMgrRelation smgr, ForkNumber forkNum,
BlockNumber blockNum, bool *foundPtr);
extern void MarkLocalBufferDirty(Buffer buffer);
+extern bool IsLocalBufferDirty(Buffer buffer);
extern void DropRelFileNodeLocalBuffers(RelFileNode rnode, ForkNumber forkNum,
BlockNumber firstDelBlock);
extern void DropRelFileNodeAllLocalBuffers(RelFileNode rnode);
diff --git a/src/include/storage/bufmgr.h b/src/include/storage/bufmgr.hindex 3cce3906a0..dbf3843053 100644--- a/src/include/storage/bufmgr.h+++ b/src/include/storage/bufmgr.h@@ -43,8 +43,10 @@ typedef enum
RBM_ZERO_AND_CLEANUP_LOCK, /* Like RBM_ZERO_AND_LOCK, but locks the page
* in "cleanup" mode */
RBM_ZERO_ON_ERROR, /* Read, but return an all-zeros page on error */
- RBM_NORMAL_NO_LOG /* Don't log page as invalid during WAL+ RBM_NORMAL_NO_LOG, /* Don't log page as invalid during WAL
* replay; otherwise same as RBM_NORMAL */
+ RBM_NORMAL_NO_READ /* Don't actually read page, returns only if it+ presents in shared memory */
} ReadBufferMode;
/* forward declared, to avoid having to expose buf_internals.h here */
@@ -175,6 +177,7 @@ extern Buffer ReadBufferWithoutRelcache(RelFileNode rnode,
extern void ReleaseBuffer(Buffer buffer);
extern void UnlockReleaseBuffer(Buffer buffer);
extern void MarkBufferDirty(Buffer buffer);
+extern bool IsBufferDirty(Buffer buffer);
extern void IncrBufferRefCount(Buffer buffer);
extern Buffer ReleaseAndReadBuffer(Buffer buffer, Relation relation,
BlockNumber blockNum);
diff --git a/src/include/storage/pmsignal.h b/src/include/storage/pmsignal.hindex 5ecc1b757c..4c73acd62c 100644--- a/src/include/storage/pmsignal.h+++ b/src/include/storage/pmsignal.h@@ -36,6 +36,8 @@ typedef enum
PMSIGNAL_ROTATE_LOGFILE, /* send SIGUSR1 to syslogger to rotate logfile */
PMSIGNAL_START_AUTOVAC_LAUNCHER, /* start an autovacuum launcher */
PMSIGNAL_START_AUTOVAC_WORKER, /* start an autovacuum worker */
+ PMSIGNAL_START_HEAPCLNR_LAUNCHER,+ PMSIGNAL_START_HEAPCLNR_WORKER,
PMSIGNAL_BACKGROUND_WORKER_CHANGE, /* background worker state change */
PMSIGNAL_START_WALRECEIVER, /* start a walreceiver */
PMSIGNAL_ADVANCE_STATE_MACHINE, /* advance postmaster's state machine */
diff --git a/src/include/utils/shash.h b/src/include/utils/shash.hnew file mode 100644index 0000000000..7a209cf5dc--- /dev/null+++ b/src/include/utils/shash.h@@ -0,0 +1,54 @@+#ifndef SHASH_H_+#define SHASH_H_++/* hash_search operations */+typedef enum+{+ SHASH_FIND,+ SHASH_ENTER,+ SHASH_REMOVE+} SHASHACTION;++/* States of a hash table element */+typedef enum+{+ SHASH_NUSED = 0,+ SHASH_USED,+ SHASH_REMOVED+} HESTATE;++typedef uint64 (*SHashValueFunc) (void *key, uint64 size, uint64 base);+typedef bool (*CompareFunc) (void* bucket1, void* bucket2);++typedef struct SHTABCTL+{+ uint64 ElementSize;+ uint64 KeySize;+ uint64 ElementsMaxNum;+ float FillFactor;+ SHashValueFunc HashFunc;+ CompareFunc CompFunc;+} SHTABCTL;++typedef struct SHTAB+{+ SHTABCTL Header;+ char *Elements;+ uint64 nElements;+ HESTATE *state; /* State of an element of hash table*/+ uint64 SeqScanCurElem;+ uint64 HTableSize;+} SHTAB;++typedef struct SHTAB* PSHTAB;++extern PSHTAB SHASH_Create(SHTABCTL shctl);+extern void SHASH_Clean(PSHTAB shtab);+extern void SHASH_Destroy(PSHTAB shtab);+extern uint64 SHASH_Entries(PSHTAB shtab);+extern void SHASH_SeqReset(PSHTAB shtab);+extern void* SHASH_SeqNext(PSHTAB shtab);+extern void* SHASH_Search(PSHTAB shtab, void *keyPtr, SHASHACTION action, bool *foundPtr);+extern uint64 DefaultHashValueFunc(void *key, uint64 size, uint64 base);++#endif /* SHASH_H_ */diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.outindex 744d501e31..b94ce20f68 100644--- a/src/test/regress/expected/rules.out+++ b/src/test/regress/expected/rules.out@@ -167,7 +167,7 @@ select * from rtest_v1;
insert into rtest_v1 values (2, 12);
insert into rtest_v1 values (2, 13);
-select * from rtest_v1;+select * from rtest_v1 ORDER BY a, b;
a | b
---+----
1 | 11
@@ -1821,6 +1821,18 @@ pg_stat_database_conflicts| SELECT d.oid AS datid,
pg_stat_get_db_conflict_bufferpin(d.oid) AS confl_bufferpin,
pg_stat_get_db_conflict_startup_deadlock(d.oid) AS confl_deadlock
FROM pg_database d;
+pg_stat_progress_cleaner| SELECT s.pid,+ s.datid,+ d.datname,+ s.relid,+ s.param1 AS relations,+ s.param2 AS queue_len,+ s.param3 AS false_hits,+ s.param4 AS cleaned_tuples,+ s.param5 AS vainly_cleaned_tuples,+ s.param6 AS missed_tuples+ FROM (pg_stat_get_progress_info('CLEANER'::text) s(pid, datid, relid, param1, param2, param3, param4, param5, param6, param7, param8, param9, param10)+ LEFT JOIN pg_database d ON ((s.datid = d.oid)));
pg_stat_progress_vacuum| SELECT s.pid,
s.datid,
d.datname,
diff --git a/src/test/regress/expected/triggers.out b/src/test/regress/expected/triggers.outindex e57c6e1c42..b9c0af7bce 100644--- a/src/test/regress/expected/triggers.out+++ b/src/test/regress/expected/triggers.out@@ -637,7 +637,7 @@ select * from trigtest2;
-- ensure we still insert, even when all triggers are disabled
insert into trigtest default values;
-select * from trigtest;+select * from trigtest ORDER BY i;
i
---
3
diff --git a/src/test/regress/input/constraints.source b/src/test/regress/input/constraints.sourceindex 98dd4210e9..2968a1230b 100644--- a/src/test/regress/input/constraints.source+++ b/src/test/regress/input/constraints.source@@ -337,7 +337,7 @@ ROLLBACK;
-- check is done at end of statement, so this should succeed
UPDATE unique_tbl SET i = i+1;
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
-- explicitly defer the constraint
BEGIN;
@@ -349,7 +349,7 @@ DELETE FROM unique_tbl WHERE t = 'tree'; -- makes constraint valid again
COMMIT; -- should succeed
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
-- try adding an initially deferred constraint
ALTER TABLE unique_tbl DROP CONSTRAINT unique_tbl_i_key;
@@ -367,7 +367,7 @@ DELETE FROM unique_tbl WHERE i = 5 AND t = 'five';
COMMIT;
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
-- should fail at commit-time
BEGIN;
@@ -420,7 +420,7 @@ UPDATE unique_tbl SET t = 'THREE' WHERE i = 3 AND t = 'Three';
COMMIT; -- should fail
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
-- test a HOT update that modifies the newly inserted tuple,
-- but should succeed because we then remove the other conflicting tuple.
@@ -431,11 +431,11 @@ INSERT INTO unique_tbl VALUES(3, 'tree'); -- should succeed for now
UPDATE unique_tbl SET t = 'threex' WHERE t = 'tree';
DELETE FROM unique_tbl WHERE t = 'three';
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
COMMIT;
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
DROP TABLE unique_tbl;
diff --git a/src/test/regress/output/constraints.source b/src/test/regress/output/constraints.sourceindex a6a1df18e7..a2ae6df587 100644--- a/src/test/regress/output/constraints.source+++ b/src/test/regress/output/constraints.source@@ -478,7 +478,7 @@ DETAIL: Key (i)=(1) already exists.
ROLLBACK;
-- check is done at end of statement, so this should succeed
UPDATE unique_tbl SET i = i+1;
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
i | t
---+------
1 | one
@@ -494,14 +494,14 @@ SET CONSTRAINTS unique_tbl_i_key DEFERRED;
INSERT INTO unique_tbl VALUES (3, 'three');
DELETE FROM unique_tbl WHERE t = 'tree'; -- makes constraint valid again
COMMIT; -- should succeed
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
i | t
---+-------
1 | one
2 | two
+ 3 | three
4 | four
5 | five
- 3 | three
(5 rows)
-- try adding an initially deferred constraint
@@ -516,14 +516,14 @@ UPDATE unique_tbl SET i = 2 WHERE i = 4 AND t = 'four';
DELETE FROM unique_tbl WHERE i = 1 AND t = 'one';
DELETE FROM unique_tbl WHERE i = 5 AND t = 'five';
COMMIT;
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
i | t
---+-------
- 3 | three
1 | five
- 5 | one- 4 | two
2 | four
+ 3 | three+ 4 | two+ 5 | one
(5 rows)
-- should fail at commit-time
@@ -581,14 +581,14 @@ UPDATE unique_tbl SET t = 'THREE' WHERE i = 3 AND t = 'Three';
COMMIT; -- should fail
ERROR: duplicate key value violates unique constraint "unique_tbl_i_key"
DETAIL: Key (i)=(3) already exists.
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
i | t
---+-------
- 3 | three
1 | five
- 5 | one- 4 | two
2 | four
+ 3 | three+ 4 | two+ 5 | one
(5 rows)
-- test a HOT update that modifies the newly inserted tuple,
@@ -597,25 +597,25 @@ BEGIN;
INSERT INTO unique_tbl VALUES(3, 'tree'); -- should succeed for now
UPDATE unique_tbl SET t = 'threex' WHERE t = 'tree';
DELETE FROM unique_tbl WHERE t = 'three';
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
i | t
---+--------
1 | five
- 5 | one- 4 | two
2 | four
3 | threex
+ 4 | two+ 5 | one
(5 rows)
COMMIT;
-SELECT * FROM unique_tbl;+SELECT * FROM unique_tbl ORDER BY i;
i | t
---+--------
1 | five
- 5 | one- 4 | two
2 | four
3 | threex
+ 4 | two+ 5 | one
(5 rows)
DROP TABLE unique_tbl;
diff --git a/src/test/regress/sql/rules.sql b/src/test/regress/sql/rules.sqlindex 3ca4c07356..c540f12a29 100644--- a/src/test/regress/sql/rules.sql+++ b/src/test/regress/sql/rules.sql@@ -181,7 +181,7 @@ delete from rtest_v1 where b = 12;
select * from rtest_v1;
insert into rtest_v1 values (2, 12);
insert into rtest_v1 values (2, 13);
-select * from rtest_v1;+select * from rtest_v1 ORDER BY a, b;
** Remember the delete rule on rtest_v1: It says
** DO INSTEAD DELETE FROM rtest_t1 WHERE a = old.a
** So this time both rows with a = 2 must get deleted
diff --git a/src/test/regress/sql/triggers.sql b/src/test/regress/sql/triggers.sqlindex 1250cd63e0..5a45fbad41 100644--- a/src/test/regress/sql/triggers.sql+++ b/src/test/regress/sql/triggers.sql@@ -419,7 +419,7 @@ delete from trigtest where i=1;
select * from trigtest2;
-- ensure we still insert, even when all triggers are disabled
insert into trigtest default values;
-select * from trigtest;+select * from trigtest ORDER BY i;
drop table trigtest2;
drop table trigtest;
--
2.17.1
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-hackers@postgresql.org
Cc: a.lepikhov@postgrespro.ru, sawada.mshk@gmail.com
Subject: Re: [WIP] [B-Tree] Retail IndexTuple deletion
In-Reply-To: <247f4455-1ca4-b246-4ebf-283fca14ce03@postgrespro.ru>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
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