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

Commitb95ae32

Browse files
committed
Avoid WAL-logging individual tuple insertions during CREATE TABLE AS
(a/k/a SELECT INTO). Instead, flush and fsync the whole relation beforecommitting. We do still need the WAL log when PITR is active, however.Simon Riggs and Tom Lane.
1 parent1bfdd1a commitb95ae32

File tree

9 files changed

+108
-27
lines changed

9 files changed

+108
-27
lines changed

‎src/backend/access/heap/heapam.c‎

Lines changed: 24 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.194 2005/06/08 15:50:21 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.195 2005/06/20 18:37:01 tgl Exp $
1212
*
1313
*
1414
* INTERFACE ROUTINES
@@ -1034,9 +1034,20 @@ heap_get_latest_tid(Relation relation,
10341034
*
10351035
* The new tuple is stamped with current transaction ID and the specified
10361036
* command ID.
1037+
*
1038+
* If use_wal is false, the new tuple is not logged in WAL, even for a
1039+
* non-temp relation. Safe usage of this behavior requires that we arrange
1040+
* that all new tuples go into new pages not containing any tuples from other
1041+
* transactions, that the relation gets fsync'd before commit, and that the
1042+
* transaction emits at least one WAL record to ensure RecordTransactionCommit
1043+
* will decide to WAL-log the commit.
1044+
*
1045+
* use_fsm is passed directly to RelationGetBufferForTuple, which see for
1046+
* more info.
10371047
*/
10381048
Oid
1039-
heap_insert(Relationrelation,HeapTupletup,CommandIdcid)
1049+
heap_insert(Relationrelation,HeapTupletup,CommandIdcid,
1050+
booluse_wal,booluse_fsm)
10401051
{
10411052
TransactionIdxid=GetCurrentTransactionId();
10421053
Bufferbuffer;
@@ -1086,7 +1097,8 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid)
10861097
heap_tuple_toast_attrs(relation,tup,NULL);
10871098

10881099
/* Find buffer to insert this tuple into */
1089-
buffer=RelationGetBufferForTuple(relation,tup->t_len,InvalidBuffer);
1100+
buffer=RelationGetBufferForTuple(relation,tup->t_len,
1101+
InvalidBuffer,use_fsm);
10901102

10911103
/* NO EREPORT(ERROR) from here till changes are logged */
10921104
START_CRIT_SECTION();
@@ -1096,7 +1108,12 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid)
10961108
pgstat_count_heap_insert(&relation->pgstat_info);
10971109

10981110
/* XLOG stuff */
1099-
if (!relation->rd_istemp)
1111+
if (relation->rd_istemp)
1112+
{
1113+
/* No XLOG record, but still need to flag that XID exists on disk */
1114+
MyXactMadeTempRelUpdate= true;
1115+
}
1116+
elseif (use_wal)
11001117
{
11011118
xl_heap_insertxlrec;
11021119
xl_heap_headerxlhdr;
@@ -1151,11 +1168,6 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid)
11511168
PageSetLSN(page,recptr);
11521169
PageSetTLI(page,ThisTimeLineID);
11531170
}
1154-
else
1155-
{
1156-
/* No XLOG record, but still need to flag that XID exists on disk */
1157-
MyXactMadeTempRelUpdate= true;
1158-
}
11591171

11601172
END_CRIT_SECTION();
11611173

@@ -1183,7 +1195,7 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid)
11831195
Oid
11841196
simple_heap_insert(Relationrelation,HeapTupletup)
11851197
{
1186-
returnheap_insert(relation,tup,GetCurrentCommandId());
1198+
returnheap_insert(relation,tup,GetCurrentCommandId(), true, true);
11871199
}
11881200

11891201
/*
@@ -1743,7 +1755,7 @@ heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
17431755
{
17441756
/* Assume there's no chance to put newtup on same page. */
17451757
newbuf=RelationGetBufferForTuple(relation,newtup->t_len,
1746-
buffer);
1758+
buffer, true);
17471759
}
17481760
else
17491761
{
@@ -1760,7 +1772,7 @@ heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
17601772
*/
17611773
LockBuffer(buffer,BUFFER_LOCK_UNLOCK);
17621774
newbuf=RelationGetBufferForTuple(relation,newtup->t_len,
1763-
buffer);
1775+
buffer, true);
17641776
}
17651777
else
17661778
{

‎src/backend/access/heap/hio.c‎

Lines changed: 24 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/heap/hio.c,v 1.56 2005/05/07 21:32:23 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/heap/hio.c,v 1.57 2005/06/20 18:37:01 tgl Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -79,12 +79,26 @@ RelationPutHeapTuple(Relation relation,
7979
*happen if space is freed in that page after heap_update finds there's not
8080
*enough there).In that case, the page will be pinned and locked only once.
8181
*
82+
*If use_fsm is true (the normal case), we use FSM to help us find free
83+
*space. If use_fsm is false, we always append a new empty page to the
84+
*end of the relation if the tuple won't fit on the current target page.
85+
*This can save some cycles when we know the relation is new and doesn't
86+
*contain useful amounts of free space.
87+
*
88+
*The use_fsm = false case is also useful for non-WAL-logged additions to a
89+
*relation, if the caller holds exclusive lock and is careful to invalidate
90+
*relation->rd_targblock before the first insertion --- that ensures that
91+
*all insertions will occur into newly added pages and not be intermixed
92+
*with tuples from other transactions. That way, a crash can't risk losing
93+
*any committed data of other transactions. (See heap_insert's comments
94+
*for additional constraints needed for safe usage of this behavior.)
95+
*
8296
*ereport(ERROR) is allowed here, so this routine *must* be called
8397
*before any (unlogged) changes are made in buffer pool.
8498
*/
8599
Buffer
86100
RelationGetBufferForTuple(Relationrelation,Sizelen,
87-
BufferotherBuffer)
101+
BufferotherBuffer,booluse_fsm)
88102
{
89103
Bufferbuffer=InvalidBuffer;
90104
PagepageHeader;
@@ -121,11 +135,14 @@ RelationGetBufferForTuple(Relation relation, Size len,
121135
* on each page that proves not to be suitable.) If the FSM has no
122136
* record of a page with enough free space, we give up and extend the
123137
* relation.
138+
*
139+
* When use_fsm is false, we either put the tuple onto the existing
140+
* target page or extend the relation.
124141
*/
125142

126143
targetBlock=relation->rd_targblock;
127144

128-
if (targetBlock==InvalidBlockNumber)
145+
if (targetBlock==InvalidBlockNumber&&use_fsm)
129146
{
130147
/*
131148
* We have no cached target page, so ask the FSM for an initial
@@ -209,6 +226,10 @@ RelationGetBufferForTuple(Relation relation, Size len,
209226
ReleaseBuffer(buffer);
210227
}
211228

229+
/* Without FSM, always fall out of the loop and extend */
230+
if (!use_fsm)
231+
break;
232+
212233
/*
213234
* Update FSM as to condition of this page, and ask for another
214235
* page to try.

‎src/backend/executor/execMain.c‎

Lines changed: 38 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -26,13 +26,14 @@
2626
*
2727
*
2828
* IDENTIFICATION
29-
* $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.249 2005/05/22 22:30:19 tgl Exp $
29+
* $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.250 2005/06/20 18:37:01 tgl Exp $
3030
*
3131
*-------------------------------------------------------------------------
3232
*/
3333
#include"postgres.h"
3434

3535
#include"access/heapam.h"
36+
#include"access/xlog.h"
3637
#include"catalog/heap.h"
3738
#include"catalog/namespace.h"
3839
#include"commands/tablecmds.h"
@@ -44,6 +45,7 @@
4445
#include"optimizer/clauses.h"
4546
#include"optimizer/var.h"
4647
#include"parser/parsetree.h"
48+
#include"storage/smgr.h"
4749
#include"utils/acl.h"
4850
#include"utils/guc.h"
4951
#include"utils/lsyscache.h"
@@ -784,6 +786,20 @@ InitPlan(QueryDesc *queryDesc, bool explainOnly)
784786
* And open the constructed table for writing.
785787
*/
786788
intoRelationDesc=heap_open(intoRelationId,AccessExclusiveLock);
789+
790+
/* use_wal off requires rd_targblock be initially invalid */
791+
Assert(intoRelationDesc->rd_targblock==InvalidBlockNumber);
792+
793+
/*
794+
* We can skip WAL-logging the insertions, unless PITR is in use.
795+
*
796+
* Note that for a non-temp INTO table, this is safe only because
797+
* we know that the catalog changes above will have been WAL-logged,
798+
* and so RecordTransactionCommit will think it needs to WAL-log the
799+
* eventual transaction commit. Else the commit might be lost, even
800+
* though all the data is safely fsync'd ...
801+
*/
802+
estate->es_into_relation_use_wal=XLogArchivingActive();
787803
}
788804

789805
estate->es_into_relation_descriptor=intoRelationDesc;
@@ -979,7 +995,22 @@ ExecEndPlan(PlanState *planstate, EState *estate)
979995
* close the "into" relation if necessary, again keeping lock
980996
*/
981997
if (estate->es_into_relation_descriptor!=NULL)
998+
{
999+
/*
1000+
* If we skipped using WAL, and it's not a temp relation,
1001+
* we must force the relation down to disk before it's
1002+
* safe to commit the transaction. This requires forcing
1003+
* out any dirty buffers and then doing a forced fsync.
1004+
*/
1005+
if (!estate->es_into_relation_use_wal&&
1006+
!estate->es_into_relation_descriptor->rd_istemp)
1007+
{
1008+
FlushRelationBuffers(estate->es_into_relation_descriptor);
1009+
smgrimmedsync(estate->es_into_relation_descriptor->rd_smgr);
1010+
}
1011+
9821012
heap_close(estate->es_into_relation_descriptor,NoLock);
1013+
}
9831014

9841015
/*
9851016
* close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
@@ -1307,7 +1338,9 @@ ExecSelect(TupleTableSlot *slot,
13071338

13081339
tuple=ExecCopySlotTuple(slot);
13091340
heap_insert(estate->es_into_relation_descriptor,tuple,
1310-
estate->es_snapshot->curcid);
1341+
estate->es_snapshot->curcid,
1342+
estate->es_into_relation_use_wal,
1343+
false);/* never any point in using FSM */
13111344
/* we know there are no indexes to update */
13121345
heap_freetuple(tuple);
13131346
IncrAppended();
@@ -1386,7 +1419,8 @@ ExecInsert(TupleTableSlot *slot,
13861419
* insert the tuple
13871420
*/
13881421
newId=heap_insert(resultRelationDesc,tuple,
1389-
estate->es_snapshot->curcid);
1422+
estate->es_snapshot->curcid,
1423+
true, true);
13901424

13911425
IncrAppended();
13921426
(estate->es_processed)++;
@@ -2089,6 +2123,7 @@ EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
20892123
epqstate->es_result_relation_info=estate->es_result_relation_info;
20902124
epqstate->es_junkFilter=estate->es_junkFilter;
20912125
epqstate->es_into_relation_descriptor=estate->es_into_relation_descriptor;
2126+
epqstate->es_into_relation_use_wal=estate->es_into_relation_use_wal;
20922127
epqstate->es_param_list_info=estate->es_param_list_info;
20932128
if (estate->es_topPlan->nParamExec>0)
20942129
epqstate->es_param_exec_vals= (ParamExecData*)

‎src/backend/executor/execUtils.c‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/executor/execUtils.c,v 1.123 2005/04/28 21:47:12 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/executor/execUtils.c,v 1.124 2005/06/20 18:37:01 tgl Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -186,7 +186,9 @@ CreateExecutorState(void)
186186
estate->es_result_relation_info=NULL;
187187

188188
estate->es_junkFilter=NULL;
189+
189190
estate->es_into_relation_descriptor=NULL;
191+
estate->es_into_relation_use_wal= false;
190192

191193
estate->es_param_list_info=NULL;
192194
estate->es_param_exec_vals=NULL;

‎src/backend/storage/smgr/md.c‎

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/storage/smgr/md.c,v 1.115 2005/05/29 04:23:05 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/storage/smgr/md.c,v 1.116 2005/06/20 18:37:01 tgl Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -660,6 +660,9 @@ mdtruncate(SMgrRelation reln, BlockNumber nblocks, bool isTemp)
660660

661661
/*
662662
*mdimmedsync() -- Immediately sync a relation to stable storage.
663+
*
664+
* Note that only writes already issued are synced; this routine knows
665+
* nothing of dirty buffers that may exist inside the buffer manager.
663666
*/
664667
bool
665668
mdimmedsync(SMgrRelationreln)

‎src/backend/storage/smgr/smgr.c‎

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@
1111
*
1212
*
1313
* IDENTIFICATION
14-
* $PostgreSQL: pgsql/src/backend/storage/smgr/smgr.c,v 1.90 2005/06/17 22:32:46 tgl Exp $
14+
* $PostgreSQL: pgsql/src/backend/storage/smgr/smgr.c,v 1.91 2005/06/20 18:37:01 tgl Exp $
1515
*
1616
*-------------------------------------------------------------------------
1717
*/
@@ -650,7 +650,8 @@ smgrtruncate(SMgrRelation reln, BlockNumber nblocks, bool isTemp)
650650
/*
651651
*smgrimmedsync() -- Force the specified relation to stable storage.
652652
*
653-
*Synchronously force all of the specified relation down to disk.
653+
*Synchronously force all previous writes to the specified relation
654+
*down to disk.
654655
*
655656
*This is useful for building completely new relations (eg, new
656657
*indexes). Instead of incrementally WAL-logging the index build
@@ -664,6 +665,10 @@ smgrtruncate(SMgrRelation reln, BlockNumber nblocks, bool isTemp)
664665
*
665666
*The preceding writes should specify isTemp = true to avoid
666667
*duplicative fsyncs.
668+
*
669+
*Note that you need to do FlushRelationBuffers() first if there is
670+
*any possibility that there are dirty buffers for the relation;
671+
*otherwise the sync is not very meaningful.
667672
*/
668673
void
669674
smgrimmedsync(SMgrRelationreln)

‎src/include/access/heapam.h‎

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
10-
* $PostgreSQL: pgsql/src/include/access/heapam.h,v 1.101 2005/06/06 17:01:24 tgl Exp $
10+
* $PostgreSQL: pgsql/src/include/access/heapam.h,v 1.102 2005/06/20 18:37:01 tgl Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -156,7 +156,8 @@ extern ItemPointer heap_get_latest_tid(Relation relation, Snapshot snapshot,
156156
ItemPointertid);
157157
externvoidsetLastTid(constItemPointertid);
158158

159-
externOidheap_insert(Relationrelation,HeapTupletup,CommandIdcid);
159+
externOidheap_insert(Relationrelation,HeapTupletup,CommandIdcid,
160+
booluse_wal,booluse_fsm);
160161
externHTSU_Resultheap_delete(Relationrelation,ItemPointertid,ItemPointerctid,
161162
CommandIdcid,Snapshotcrosscheck,boolwait);
162163
externHTSU_Resultheap_update(Relationrelation,ItemPointerotid,HeapTupletup,

‎src/include/access/hio.h‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
10-
* $PostgreSQL: pgsql/src/include/access/hio.h,v 1.27 2004/12/31 22:03:21 pgsql Exp $
10+
* $PostgreSQL: pgsql/src/include/access/hio.h,v 1.28 2005/06/20 18:37:01 tgl Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -19,6 +19,6 @@
1919
externvoidRelationPutHeapTuple(Relationrelation,Bufferbuffer,
2020
HeapTupletuple);
2121
externBufferRelationGetBufferForTuple(Relationrelation,Sizelen,
22-
BufferotherBuffer);
22+
BufferotherBuffer,booluse_fsm);
2323

2424
#endif/* HIO_H */

‎src/include/nodes/execnodes.h‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
10-
* $PostgreSQL: pgsql/src/include/nodes/execnodes.h,v 1.134 2005/06/15 07:27:44 neilc Exp $
10+
* $PostgreSQL: pgsql/src/include/nodes/execnodes.h,v 1.135 2005/06/20 18:37:02 tgl Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -304,7 +304,9 @@ typedef struct EState
304304
ResultRelInfo*es_result_relation_info;/* currently active array
305305
* elt */
306306
JunkFilter*es_junkFilter;/* currently active junk filter */
307+
307308
Relationes_into_relation_descriptor;/* for SELECT INTO */
309+
booles_into_relation_use_wal;
308310

309311
/* Parameter info: */
310312
ParamListInfoes_param_list_info;/* values of external params */

0 commit comments

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