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

Commite981653

Browse files
committed
Update FSM on WAL replay. This is a bit limited; the FSM is only updated
on non-full-page-image WAL records, and quite arbitrarily, only if there'sless than 20% free space on the page after the insert/update (not on HOTupdates, though). The 20% cutoff should avoid most of the overhead, whenreplaying a bulk insertion, for example, while ensuring that pages thatare full are marked as full in the FSM.This is mostly to avoid the nasty worst case scenario, where you replayfrom a PITR archive, and the FSM information in the base backup is reallyout of date. If there was a lot of pages that the outdated FSM claims tohave free space, but don't actually have any, the first unlucky inserterafter the recovery would traverse through all those pages, just to findout that they're full. We didn't have this problem with the old FSMimplementation, because we simply threw the FSM information away on anon-clean shutdown.
1 parent9b46abb commite981653

File tree

3 files changed

+92
-9
lines changed

3 files changed

+92
-9
lines changed

‎src/backend/access/heap/heapam.c

Lines changed: 58 additions & 7 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 1.267 2008/10/3115:04:59 heikki Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.268 2008/10/3119:40:26 heikki Exp $
1212
*
1313
*
1414
* INTERFACE ROUTINES
@@ -54,6 +54,7 @@
5454
#include"miscadmin.h"
5555
#include"pgstat.h"
5656
#include"storage/bufmgr.h"
57+
#include"storage/freespace.h"
5758
#include"storage/lmgr.h"
5859
#include"storage/procarray.h"
5960
#include"storage/smgr.h"
@@ -4022,6 +4023,7 @@ heap_xlog_clean(XLogRecPtr lsn, XLogRecord *record, bool clean_move)
40224023
intnredirected;
40234024
intndead;
40244025
intnunused;
4026+
Sizefreespace;
40254027

40264028
if (record->xl_info&XLR_BKP_BLOCK_1)
40274029
return;
@@ -4053,6 +4055,8 @@ heap_xlog_clean(XLogRecPtr lsn, XLogRecord *record, bool clean_move)
40534055
nowunused,nunused,
40544056
clean_move);
40554057

4058+
freespace=PageGetHeapFreeSpace(page);/* needed to update FSM below */
4059+
40564060
/*
40574061
* Note: we don't worry about updating the page's prunability hints.
40584062
* At worst this will cause an extra prune cycle to occur soon.
@@ -4062,6 +4066,15 @@ heap_xlog_clean(XLogRecPtr lsn, XLogRecord *record, bool clean_move)
40624066
PageSetTLI(page,ThisTimeLineID);
40634067
MarkBufferDirty(buffer);
40644068
UnlockReleaseBuffer(buffer);
4069+
4070+
/*
4071+
* Update the FSM as well.
4072+
*
4073+
* XXX: We don't get here if the page was restored from full page image.
4074+
* We don't bother to update the FSM in that case, it doesn't need to be
4075+
* totally accurate anyway.
4076+
*/
4077+
XLogRecordPageWithFreeSpace(xlrec->node,xlrec->block,freespace);
40654078
}
40664079

40674080
staticvoid
@@ -4205,25 +4218,25 @@ heap_xlog_insert(XLogRecPtr lsn, XLogRecord *record)
42054218
HeapTupleHeaderhtup;
42064219
xl_heap_headerxlhdr;
42074220
uint32newlen;
4221+
Sizefreespace;
4222+
BlockNumberblkno;
42084223

42094224
if (record->xl_info&XLR_BKP_BLOCK_1)
42104225
return;
42114226

4227+
blkno=ItemPointerGetBlockNumber(&(xlrec->target.tid));
4228+
42124229
if (record->xl_info&XLOG_HEAP_INIT_PAGE)
42134230
{
4214-
buffer=XLogReadBuffer(xlrec->target.node,
4215-
ItemPointerGetBlockNumber(&(xlrec->target.tid)),
4216-
true);
4231+
buffer=XLogReadBuffer(xlrec->target.node,blkno, true);
42174232
Assert(BufferIsValid(buffer));
42184233
page= (Page)BufferGetPage(buffer);
42194234

42204235
PageInit(page,BufferGetPageSize(buffer),0);
42214236
}
42224237
else
42234238
{
4224-
buffer=XLogReadBuffer(xlrec->target.node,
4225-
ItemPointerGetBlockNumber(&(xlrec->target.tid)),
4226-
false);
4239+
buffer=XLogReadBuffer(xlrec->target.node,blkno, false);
42274240
if (!BufferIsValid(buffer))
42284241
return;
42294242
page= (Page)BufferGetPage(buffer);
@@ -4261,10 +4274,25 @@ heap_xlog_insert(XLogRecPtr lsn, XLogRecord *record)
42614274
offnum=PageAddItem(page, (Item)htup,newlen,offnum, true, true);
42624275
if (offnum==InvalidOffsetNumber)
42634276
elog(PANIC,"heap_insert_redo: failed to add tuple");
4277+
4278+
freespace=PageGetHeapFreeSpace(page);/* needed to update FSM below */
4279+
42644280
PageSetLSN(page,lsn);
42654281
PageSetTLI(page,ThisTimeLineID);
42664282
MarkBufferDirty(buffer);
42674283
UnlockReleaseBuffer(buffer);
4284+
4285+
/*
4286+
* If the page is running low on free space, update the FSM as well.
4287+
* Arbitrarily, our definition of "low" is less than 20%. We can't do
4288+
* much better than that without knowing the fill-factor for the table.
4289+
*
4290+
* XXX: We don't get here if the page was restored from full page image.
4291+
* We don't bother to update the FSM in that case, it doesn't need to be
4292+
* totally accurate anyway.
4293+
*/
4294+
if (freespace<BLCKSZ /5)
4295+
XLogRecordPageWithFreeSpace(xlrec->target.node,blkno,freespace);
42684296
}
42694297

42704298
/*
@@ -4289,6 +4317,7 @@ heap_xlog_update(XLogRecPtr lsn, XLogRecord *record, bool move, bool hot_update)
42894317
xl_heap_headerxlhdr;
42904318
inthsize;
42914319
uint32newlen;
4320+
Sizefreespace;
42924321

42934322
if (record->xl_info&XLR_BKP_BLOCK_1)
42944323
{
@@ -4446,10 +4475,32 @@ newsame:;
44464475
offnum=PageAddItem(page, (Item)htup,newlen,offnum, true, true);
44474476
if (offnum==InvalidOffsetNumber)
44484477
elog(PANIC,"heap_update_redo: failed to add tuple");
4478+
4479+
freespace=PageGetHeapFreeSpace(page);/* needed to update FSM below */
4480+
44494481
PageSetLSN(page,lsn);
44504482
PageSetTLI(page,ThisTimeLineID);
44514483
MarkBufferDirty(buffer);
44524484
UnlockReleaseBuffer(buffer);
4485+
4486+
/*
4487+
* If the page is running low on free space, update the FSM as well.
4488+
* Arbitrarily, our definition of "low" is less than 20%. We can't do
4489+
* much better than that without knowing the fill-factor for the table.
4490+
*
4491+
* However, don't update the FSM on HOT updates, because after crash
4492+
* recovery, either the old or the new tuple will certainly be dead and
4493+
* prunable. After pruning, the page will have roughly as much free space
4494+
* as it did before the update, assuming the new tuple is about the same
4495+
* size as the old one.
4496+
*
4497+
* XXX: We don't get here if the page was restored from full page image.
4498+
* We don't bother to update the FSM in that case, it doesn't need to be
4499+
* totally accurate anyway.
4500+
*/
4501+
if (!hot_update&&freespace<BLCKSZ /5)
4502+
XLogRecordPageWithFreeSpace(xlrec->target.node,
4503+
ItemPointerGetBlockNumber(&(xlrec->newtid)),freespace);
44534504
}
44544505

44554506
staticvoid

‎src/backend/storage/freespace/freespace.c

Lines changed: 31 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/storage/freespace/freespace.c,v 1.65 2008/10/3115:05:00 heikki Exp $
11+
* $PostgreSQL: pgsql/src/backend/storage/freespace/freespace.c,v 1.66 2008/10/3119:40:27 heikki Exp $
1212
*
1313
*
1414
* NOTES:
@@ -202,6 +202,36 @@ RecordPageWithFreeSpace(Relation rel, BlockNumber heapBlk, Size spaceAvail)
202202
fsm_set_and_search(rel,addr,slot,new_cat,0);
203203
}
204204

205+
/*
206+
* XLogRecordPageWithFreeSpace - like RecordPageWithFreeSpace, for use in
207+
*WAL replay
208+
*/
209+
void
210+
XLogRecordPageWithFreeSpace(RelFileNodernode,BlockNumberheapBlk,
211+
SizespaceAvail)
212+
{
213+
intnew_cat=fsm_space_avail_to_cat(spaceAvail);
214+
FSMAddressaddr;
215+
uint16slot;
216+
BlockNumberblkno;
217+
Bufferbuf;
218+
Pagepage;
219+
220+
/* Get the location of the FSM byte representing the heap block */
221+
addr=fsm_get_location(heapBlk,&slot);
222+
blkno=fsm_logical_to_physical(addr);
223+
224+
/* If the page doesn't exist already, extend */
225+
buf=XLogReadBufferExtended(rnode,FSM_FORKNUM,blkno,RBM_ZERO_ON_ERROR);
226+
page=BufferGetPage(buf);
227+
if (PageIsNew(page))
228+
PageInit(page,BLCKSZ,0);
229+
230+
if (fsm_set_avail(page,slot,new_cat))
231+
MarkBufferDirty(buf);
232+
UnlockReleaseBuffer(buf);
233+
}
234+
205235
/*
206236
* GetRecordedFreePage - return the amount of free space on a particular page,
207237
*according to the FSM.

‎src/include/storage/freespace.h

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
* Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
10-
* $PostgreSQL: pgsql/src/include/storage/freespace.h,v 1.29 2008/09/30 10:52:13 heikki Exp $
10+
* $PostgreSQL: pgsql/src/include/storage/freespace.h,v 1.30 2008/10/31 19:40:27 heikki Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -27,6 +27,8 @@ extern BlockNumber RecordAndGetPageWithFreeSpace(Relation rel,
2727
SizespaceNeeded);
2828
externvoidRecordPageWithFreeSpace(Relationrel,BlockNumberheapBlk,
2929
SizespaceAvail);
30+
externvoidXLogRecordPageWithFreeSpace(RelFileNodernode,BlockNumberheapBlk,
31+
SizespaceAvail);
3032

3133
externvoidFreeSpaceMapTruncateRel(Relationrel,BlockNumbernblocks);
3234
externvoidFreeSpaceMapVacuum(Relationrel);

0 commit comments

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