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

Commit0454f13

Browse files
committed
Rewrite the rbtree routines so that an RBNode is the first field of the
struct representing a tree entry, rather than being a separately allocatedpiece of storage. This API is at least as clean as the old one (if notmore so --- there were some bizarre choices in there) and it permits avery substantial memory savings, on the order of 2X in ginbulk.c's usage.Also, fix minor memory leaks in code called by ginEntryInsert, inparticular in ginInsertValue and entryFillRoot, as well as ginEntryInsertitself. These leaks resulted in the GIN index build context continuingto bloat even after we'd filled it to maintenance_work_mem and startedto dump data out to the index.In combination these fixes restore the GIN index build code to honoringthe maintenance_work_mem limit about as well as it did in 8.4. Speedseems on par with 8.4 too, maybe even a bit faster, for a non-pathologicalcase in which HEAD was formerly slower.Back-patch to 9.0 so we don't have a performance regression from 8.4.
1 parentafc2900 commit0454f13

File tree

8 files changed

+410
-293
lines changed

8 files changed

+410
-293
lines changed

‎src/backend/access/gin/ginbtree.c

Lines changed: 9 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
1010
* IDENTIFICATION
11-
*$PostgreSQL: pgsql/src/backend/access/gin/ginbtree.c,v 1.15 2010/01/02 16:57:33 momjian Exp $
11+
*$PostgreSQL: pgsql/src/backend/access/gin/ginbtree.c,v 1.16 2010/08/01 02:12:42 tgl Exp $
1212
*-------------------------------------------------------------------------
1313
*/
1414

@@ -267,6 +267,8 @@ findParents(GinBtree btree, GinBtreeStack *stack,
267267

268268
/*
269269
* Insert value (stored in GinBtree) to tree described by stack
270+
*
271+
* NB: the passed-in stack is freed, as though by freeGinBtreeStack.
270272
*/
271273
void
272274
ginInsertValue(GinBtreebtree,GinBtreeStack*stack)
@@ -308,10 +310,11 @@ ginInsertValue(GinBtree btree, GinBtreeStack *stack)
308310
PageSetTLI(page,ThisTimeLineID);
309311
}
310312

311-
UnlockReleaseBuffer(stack->buffer);
313+
LockBuffer(stack->buffer,GIN_UNLOCK);
312314
END_CRIT_SECTION();
313315

314-
freeGinBtreeStack(stack->parent);
316+
freeGinBtreeStack(stack);
317+
315318
return;
316319
}
317320
else
@@ -325,7 +328,6 @@ ginInsertValue(GinBtree btree, GinBtreeStack *stack)
325328
*/
326329
newlpage=btree->splitPage(btree,stack->buffer,rbuffer,stack->off,&rdata);
327330

328-
329331
((ginxlogSplit*) (rdata->data))->rootBlkno=rootBlkno;
330332

331333
parent=stack->parent;
@@ -341,7 +343,6 @@ ginInsertValue(GinBtree btree, GinBtreeStack *stack)
341343
((ginxlogSplit*) (rdata->data))->isRootSplit= TRUE;
342344
((ginxlogSplit*) (rdata->data))->rrlink=InvalidBlockNumber;
343345

344-
345346
page=BufferGetPage(stack->buffer);
346347
lpage=BufferGetPage(lbuffer);
347348
rpage=BufferGetPage(rbuffer);
@@ -375,10 +376,11 @@ ginInsertValue(GinBtree btree, GinBtreeStack *stack)
375376

376377
UnlockReleaseBuffer(rbuffer);
377378
UnlockReleaseBuffer(lbuffer);
378-
UnlockReleaseBuffer(stack->buffer);
379-
379+
LockBuffer(stack->buffer,GIN_UNLOCK);
380380
END_CRIT_SECTION();
381381

382+
freeGinBtreeStack(stack);
383+
382384
return;
383385
}
384386
else

‎src/backend/access/gin/ginbulk.c

Lines changed: 78 additions & 56 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
1010
* IDENTIFICATION
11-
*$PostgreSQL: pgsql/src/backend/access/gin/ginbulk.c,v 1.19 2010/02/26 02:00:33 momjian Exp $
11+
*$PostgreSQL: pgsql/src/backend/access/gin/ginbulk.c,v 1.20 2010/08/01 02:12:42 tgl Exp $
1212
*-------------------------------------------------------------------------
1313
*/
1414

@@ -19,17 +19,21 @@
1919
#include"utils/memutils.h"
2020

2121

22-
#defineDEF_NENTRY2048
23-
#defineDEF_NPTR4
22+
#defineDEF_NENTRY2048/* EntryAccumulator allocation quantum */
23+
#defineDEF_NPTR5/* ItemPointer initial allocation quantum */
2424

25-
staticvoid*
26-
ginAppendData(void*old,void*new,void*arg)
27-
{
28-
EntryAccumulator*eo= (EntryAccumulator*)old,
29-
*en= (EntryAccumulator*)new;
3025

26+
/* Combiner function for rbtree.c */
27+
staticvoid
28+
ginCombineData(RBNode*existing,constRBNode*newdata,void*arg)
29+
{
30+
EntryAccumulator*eo= (EntryAccumulator*)existing;
31+
constEntryAccumulator*en= (constEntryAccumulator*)newdata;
3132
BuildAccumulator*accum= (BuildAccumulator*)arg;
3233

34+
/*
35+
* Note this code assumes that newdata contains only one itempointer.
36+
*/
3337
if (eo->number >=eo->length)
3438
{
3539
accum->allocatedMemory-=GetMemoryChunkSpace(eo->list);
@@ -53,29 +57,57 @@ ginAppendData(void *old, void *new, void *arg)
5357

5458
eo->list[eo->number]=en->list[0];
5559
eo->number++;
56-
57-
returnold;
5860
}
5961

62+
/* Comparator function for rbtree.c */
6063
staticint
61-
cmpEntryAccumulator(constvoid*a,constvoid*b,void*arg)
64+
cmpEntryAccumulator(constRBNode*a,constRBNode*b,void*arg)
6265
{
63-
EntryAccumulator*ea= (EntryAccumulator*)a;
64-
EntryAccumulator*eb= (EntryAccumulator*)b;
66+
constEntryAccumulator*ea= (constEntryAccumulator*)a;
67+
constEntryAccumulator*eb= (constEntryAccumulator*)b;
6568
BuildAccumulator*accum= (BuildAccumulator*)arg;
6669

6770
returncompareAttEntries(accum->ginstate,ea->attnum,ea->value,
6871
eb->attnum,eb->value);
6972
}
7073

74+
/* Allocator function for rbtree.c */
75+
staticRBNode*
76+
ginAllocEntryAccumulator(void*arg)
77+
{
78+
BuildAccumulator*accum= (BuildAccumulator*)arg;
79+
EntryAccumulator*ea;
80+
81+
/*
82+
* Allocate memory by rather big chunks to decrease overhead. We have
83+
* no need to reclaim RBNodes individually, so this costs nothing.
84+
*/
85+
if (accum->entryallocator==NULL||accum->length >=DEF_NENTRY)
86+
{
87+
accum->entryallocator=palloc(sizeof(EntryAccumulator)*DEF_NENTRY);
88+
accum->allocatedMemory+=GetMemoryChunkSpace(accum->entryallocator);
89+
accum->length=0;
90+
}
91+
92+
/* Allocate new RBNode from current chunk */
93+
ea=accum->entryallocator+accum->length;
94+
accum->length++;
95+
96+
return (RBNode*)ea;
97+
}
98+
7199
void
72100
ginInitBA(BuildAccumulator*accum)
73101
{
74102
accum->allocatedMemory=0;
103+
accum->length=0;
75104
accum->entryallocator=NULL;
76-
accum->tree=rb_create(cmpEntryAccumulator,ginAppendData,NULL,accum);
77-
accum->iterator=NULL;
78-
accum->tmpList=NULL;
105+
accum->tree=rb_create(sizeof(EntryAccumulator),
106+
cmpEntryAccumulator,
107+
ginCombineData,
108+
ginAllocEntryAccumulator,
109+
NULL,/* no freefunc needed */
110+
(void*)accum);
79111
}
80112

81113
/*
@@ -104,55 +136,41 @@ getDatumCopy(BuildAccumulator *accum, OffsetNumber attnum, Datum value)
104136
staticvoid
105137
ginInsertEntry(BuildAccumulator*accum,ItemPointerheapptr,OffsetNumberattnum,Datumentry)
106138
{
107-
EntryAccumulator*key,
108-
*ea;
139+
EntryAccumulatorkey;
140+
EntryAccumulator*ea;
141+
boolisNew;
109142

110143
/*
111-
* Allocate memory by rather big chunk to decrease overhead, we don't keep
112-
* pointer to previously allocated chunks because they will free by
113-
* MemoryContextReset() call.
144+
* For the moment, fill only the fields of key that will be looked at
145+
* by cmpEntryAccumulator or ginCombineData.
114146
*/
115-
if (accum->entryallocator==NULL||accum->length >=DEF_NENTRY)
116-
{
117-
accum->entryallocator=palloc(sizeof(EntryAccumulator)*DEF_NENTRY);
118-
accum->allocatedMemory+=GetMemoryChunkSpace(accum->entryallocator);
119-
accum->length=0;
120-
}
147+
key.attnum=attnum;
148+
key.value=entry;
149+
/* temporarily set up single-entry itempointer list */
150+
key.list=heapptr;
121151

122-
/* "Allocate" new key in chunk */
123-
key=accum->entryallocator+accum->length;
124-
accum->length++;
152+
ea= (EntryAccumulator*)rb_insert(accum->tree, (RBNode*)&key,&isNew);
125153

126-
key->attnum=attnum;
127-
key->value=entry;
128-
/* To prevent multiple palloc/pfree cycles, we reuse array */
129-
if (accum->tmpList==NULL)
130-
accum->tmpList=
131-
(ItemPointerData*)palloc(sizeof(ItemPointerData)*DEF_NPTR);
132-
key->list=accum->tmpList;
133-
key->list[0]=*heapptr;
134-
135-
ea=rb_insert(accum->tree,key);
136-
137-
if (ea==NULL)
154+
if (isNew)
138155
{
139156
/*
140-
* The key has been inserted, so continue initialization.
157+
* Finish initializing new tree entry, including making permanent
158+
* copies of the datum and itempointer.
141159
*/
142-
key->value=getDatumCopy(accum,attnum,entry);
143-
key->length=DEF_NPTR;
144-
key->number=1;
145-
key->shouldSort= FALSE;
146-
accum->allocatedMemory+=GetMemoryChunkSpace(key->list);
147-
accum->tmpList=NULL;
160+
ea->value=getDatumCopy(accum,attnum,entry);
161+
ea->length=DEF_NPTR;
162+
ea->number=1;
163+
ea->shouldSort= FALSE;
164+
ea->list=
165+
(ItemPointerData*)palloc(sizeof(ItemPointerData)*DEF_NPTR);
166+
ea->list[0]=*heapptr;
167+
accum->allocatedMemory+=GetMemoryChunkSpace(ea->list);
148168
}
149169
else
150170
{
151171
/*
152-
* The key has been appended, so "free" allocated key by decrementing
153-
* chunk's counter.
172+
* ginCombineData did everything needed.
154173
*/
155-
accum->length--;
156174
}
157175
}
158176

@@ -214,16 +232,20 @@ qsortCompareItemPointers(const void *a, const void *b)
214232
returnres;
215233
}
216234

235+
/* Prepare to read out the rbtree contents using ginGetEntry */
236+
void
237+
ginBeginBAScan(BuildAccumulator*accum)
238+
{
239+
rb_begin_iterate(accum->tree,LeftRightWalk);
240+
}
241+
217242
ItemPointerData*
218243
ginGetEntry(BuildAccumulator*accum,OffsetNumber*attnum,Datum*value,uint32*n)
219244
{
220245
EntryAccumulator*entry;
221246
ItemPointerData*list;
222247

223-
if (accum->iterator==NULL)
224-
accum->iterator=rb_begin_iterate(accum->tree,LeftRightWalk);
225-
226-
entry=rb_iterate(accum->iterator);
248+
entry= (EntryAccumulator*)rb_iterate(accum->tree);
227249

228250
if (entry==NULL)
229251
returnNULL;

‎src/backend/access/gin/ginentrypage.c

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
1010
* IDENTIFICATION
11-
*$PostgreSQL: pgsql/src/backend/access/gin/ginentrypage.c,v 1.24 2010/02/26 02:00:33 momjian Exp $
11+
*$PostgreSQL: pgsql/src/backend/access/gin/ginentrypage.c,v 1.25 2010/08/01 02:12:42 tgl Exp $
1212
*-------------------------------------------------------------------------
1313
*/
1414

@@ -615,7 +615,7 @@ entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off, XLogR
615615
}
616616

617617
/*
618-
* return newlyallocate rightmost tuple
618+
* return newlyallocated rightmost tuple
619619
*/
620620
IndexTuple
621621
ginPageGetLinkItup(Bufferbuf)
@@ -646,10 +646,12 @@ entryFillRoot(GinBtree btree, Buffer root, Buffer lbuf, Buffer rbuf)
646646
itup=ginPageGetLinkItup(lbuf);
647647
if (PageAddItem(page, (Item)itup,IndexTupleSize(itup),InvalidOffsetNumber, false, false)==InvalidOffsetNumber)
648648
elog(ERROR,"failed to add item to index root page");
649+
pfree(itup);
649650

650651
itup=ginPageGetLinkItup(rbuf);
651652
if (PageAddItem(page, (Item)itup,IndexTupleSize(itup),InvalidOffsetNumber, false, false)==InvalidOffsetNumber)
652653
elog(ERROR,"failed to add item to index root page");
654+
pfree(itup);
653655
}
654656

655657
void

‎src/backend/access/gin/ginfast.c

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@
1111
* Portions Copyright (c) 1994, Regents of the University of California
1212
*
1313
* IDENTIFICATION
14-
*$PostgreSQL: pgsql/src/backend/access/gin/ginfast.c,v 1.7 2010/02/11 14:29:50 teodor Exp $
14+
*$PostgreSQL: pgsql/src/backend/access/gin/ginfast.c,v 1.8 2010/08/01 02:12:42 tgl Exp $
1515
*
1616
*-------------------------------------------------------------------------
1717
*/
@@ -786,6 +786,7 @@ ginInsertCleanup(Relation index, GinState *ginstate,
786786
* significant amount of time - so, run it without locking pending
787787
* list.
788788
*/
789+
ginBeginBAScan(&accum);
789790
while ((list=ginGetEntry(&accum,&attnum,&entry,&nlist))!=NULL)
790791
{
791792
ginEntryInsert(index,ginstate,attnum,entry,list,nlist, FALSE);
@@ -820,6 +821,7 @@ ginInsertCleanup(Relation index, GinState *ginstate,
820821
ginInitBA(&accum);
821822
processPendingPage(&accum,&datums,page,maxoff+1);
822823

824+
ginBeginBAScan(&accum);
823825
while ((list=ginGetEntry(&accum,&attnum,&entry,&nlist))!=NULL)
824826
ginEntryInsert(index,ginstate,attnum,entry,list,nlist, FALSE);
825827
}

‎src/backend/access/gin/gininsert.c

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
1010
* IDENTIFICATION
11-
*$PostgreSQL: pgsql/src/backend/access/gin/gininsert.c,v 1.26 2010/02/11 14:29:50 teodor Exp $
11+
*$PostgreSQL: pgsql/src/backend/access/gin/gininsert.c,v 1.27 2010/08/01 02:12:42 tgl Exp $
1212
*-------------------------------------------------------------------------
1313
*/
1414

@@ -176,6 +176,7 @@ ginEntryInsert(Relation index, GinState *ginstate,
176176
gdi=prepareScanPostingTree(index,rootPostingTree, FALSE);
177177
gdi->btree.isBuild=isBuild;
178178
insertItemPointer(gdi,items,nitem);
179+
pfree(gdi);
179180

180181
return;
181182
}
@@ -254,6 +255,7 @@ ginBuildCallback(Relation index, HeapTuple htup, Datum *values,
254255
uint32nlist;
255256
OffsetNumberattnum;
256257

258+
ginBeginBAScan(&buildstate->accum);
257259
while ((list=ginGetEntry(&buildstate->accum,&attnum,&entry,&nlist))!=NULL)
258260
{
259261
/* there could be many entries, so be willing to abort here */
@@ -360,6 +362,7 @@ ginbuild(PG_FUNCTION_ARGS)
360362

361363
/* dump remaining entries to the index */
362364
oldCtx=MemoryContextSwitchTo(buildstate.tmpCtx);
365+
ginBeginBAScan(&buildstate.accum);
363366
while ((list=ginGetEntry(&buildstate.accum,&attnum,&entry,&nlist))!=NULL)
364367
{
365368
/* there could be many entries, so be willing to abort here */

0 commit comments

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