Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commite034e51

Browse files
author
Neil Conway
committed
Fix a small memory leak in ExecReScanAgg() in the hashed aggregation case.
In the previous coding, the list of columns that needed to be hashed onwas allocated in the per-query context, but we reallocated every timethe Agg node was rescanned. Since this information doesn't change overa rescan, just construct the list of columns once during ExecInitAgg().
1 parent0e3d5ad commite034e51

File tree

1 file changed

+39
-28
lines changed

1 file changed

+39
-28
lines changed

‎src/backend/executor/nodeAgg.c

Lines changed: 39 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -61,7 +61,7 @@
6161
* Portions Copyright (c) 1994, Regents of the University of California
6262
*
6363
* IDENTIFICATION
64-
* $PostgreSQL: pgsql/src/backend/executor/nodeAgg.c,v 1.161 2008/09/08 00:22:55tgl Exp $
64+
* $PostgreSQL: pgsql/src/backend/executor/nodeAgg.c,v 1.162 2008/10/16 19:25:55neilc Exp $
6565
*
6666
*-------------------------------------------------------------------------
6767
*/
@@ -665,9 +665,6 @@ build_hash_table(AggState *aggstate)
665665
Agg*node= (Agg*)aggstate->ss.ps.plan;
666666
MemoryContexttmpmem=aggstate->tmpcontext->ecxt_per_tuple_memory;
667667
Sizeentrysize;
668-
Bitmapset*colnos;
669-
List*collist;
670-
inti;
671668

672669
Assert(node->aggstrategy==AGG_HASHED);
673670
Assert(node->numGroups>0);
@@ -683,30 +680,40 @@ build_hash_table(AggState *aggstate)
683680
entrysize,
684681
aggstate->aggcontext,
685682
tmpmem);
683+
}
686684

687-
/*
688-
* Create a list of the tuple columns that actually need to be stored in
689-
* hashtable entries. The incoming tuples from the child plan node will
690-
* contain grouping columns, other columns referenced in our targetlist
691-
* and qual, columns used to compute the aggregate functions, and perhaps
692-
* just junk columns we don't use at all. Only columns of the first two
693-
* types need to be stored in the hashtable, and getting rid of the others
694-
* can make the table entries significantly smaller. To avoid messing up
695-
* Var numbering, we keep the same tuple descriptor for hashtable entries
696-
* as the incoming tuples have, but set unwanted columns to NULL in the
697-
* tuples that go into the table.
698-
*
699-
* To eliminate duplicates, we build a bitmapset of the needed columns,
700-
* then convert it to an integer list (cheaper to scan at runtime). The
701-
* list is in decreasing order so that the first entry is the largest;
702-
* lookup_hash_entry depends on this to use slot_getsomeattrs correctly.
703-
*
704-
* Note: at present, searching the tlist/qual is not really necessary
705-
* since the parser should disallow any unaggregated references to
706-
* ungrouped columns. However, the search will be needed when we add
707-
* support for SQL99 semantics that allow use of "functionally dependent"
708-
* columns that haven't been explicitly grouped by.
709-
*/
685+
/*
686+
* Create a list of the tuple columns that actually need to be stored in
687+
* hashtable entries. The incoming tuples from the child plan node will
688+
* contain grouping columns, other columns referenced in our targetlist and
689+
* qual, columns used to compute the aggregate functions, and perhaps just
690+
* junk columns we don't use at all. Only columns of the first two types
691+
* need to be stored in the hashtable, and getting rid of the others can
692+
* make the table entries significantly smaller. To avoid messing up Var
693+
* numbering, we keep the same tuple descriptor for hashtable entries as the
694+
* incoming tuples have, but set unwanted columns to NULL in the tuples that
695+
* go into the table.
696+
*
697+
* To eliminate duplicates, we build a bitmapset of the needed columns, then
698+
* convert it to an integer list (cheaper to scan at runtime). The list is
699+
* in decreasing order so that the first entry is the largest;
700+
* lookup_hash_entry depends on this to use slot_getsomeattrs correctly.
701+
* Note that the list is preserved over ExecReScanAgg, so we allocate it in
702+
* the per-query context (unlike the hash table itself).
703+
*
704+
* Note: at present, searching the tlist/qual is not really necessary since
705+
* the parser should disallow any unaggregated references to ungrouped
706+
* columns. However, the search will be needed when we add support for
707+
* SQL99 semantics that allow use of "functionally dependent" columns that
708+
* haven't been explicitly grouped by.
709+
*/
710+
staticList*
711+
find_hash_columns(AggState*aggstate)
712+
{
713+
Agg*node= (Agg*)aggstate->ss.ps.plan;
714+
Bitmapset*colnos;
715+
List*collist;
716+
inti;
710717

711718
/* Find Vars that will be needed in tlist and qual */
712719
colnos=find_unaggregated_cols(aggstate);
@@ -717,7 +724,9 @@ build_hash_table(AggState *aggstate)
717724
collist=NIL;
718725
while ((i=bms_first_member(colnos)) >=0)
719726
collist=lcons_int(i,collist);
720-
aggstate->hash_needed=collist;
727+
bms_free(colnos);
728+
729+
returncollist;
721730
}
722731

723732
/*
@@ -1325,6 +1334,8 @@ ExecInitAgg(Agg *node, EState *estate, int eflags)
13251334
{
13261335
build_hash_table(aggstate);
13271336
aggstate->table_filled= false;
1337+
/* Compute the columns we actually need to hash on */
1338+
aggstate->hash_needed=find_hash_columns(aggstate);
13281339
}
13291340
else
13301341
{

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp