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

Commit1ed3c6f

Browse files
committed
Log the creation of an init fork unconditionally.
Previously, it was thought that this only needed to be done for thebenefit of possible standbys, so wal_level = minimal skipped it.But that's not safe, because during crash recovery we might replayXLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record which recursivelyremoves the directory that contains the new init fork. So log italways.The user-visible effect of this bug is that if you create a databaseor tablespace, then create an unlogged table, then crash withoutcheckpointing, then restart, accessing the table will fail, becausethe it won't have been properly reset. This commit fixes that.Michael Paquier, per a report from Konstantin Knizhnik. Wording ofthe comments per a suggestion from me.
1 parent29babe9 commit1ed3c6f

File tree

4 files changed

+38
-24
lines changed

4 files changed

+38
-24
lines changed

‎contrib/bloom/blinsert.c‎

Lines changed: 9 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -164,13 +164,18 @@ blbuildempty(Relation index)
164164
metapage= (Page)palloc(BLCKSZ);
165165
BloomFillMetapage(index,metapage);
166166

167-
/* Write the page. If archiving/streaming, XLOG it. */
167+
/*
168+
* Write the page and log it. It might seem that an immediate sync
169+
* would be sufficient to guarantee that the file exists on disk, but
170+
* recovery itself might remove it while replaying, for example, an
171+
* XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we
172+
* need this even when wal_level=minimal.
173+
*/
168174
PageSetChecksumInplace(metapage,BLOOM_METAPAGE_BLKNO);
169175
smgrwrite(index->rd_smgr,INIT_FORKNUM,BLOOM_METAPAGE_BLKNO,
170176
(char*)metapage, true);
171-
if (XLogIsNeeded())
172-
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
173-
BLOOM_METAPAGE_BLKNO,metapage, false);
177+
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
178+
BLOOM_METAPAGE_BLKNO,metapage, false);
174179

175180
/*
176181
* An immediate sync is required even if we xlog'd the page, because the

‎src/backend/access/nbtree/nbtree.c‎

Lines changed: 9 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -242,13 +242,18 @@ btbuildempty(Relation index)
242242
metapage= (Page)palloc(BLCKSZ);
243243
_bt_initmetapage(metapage,P_NONE,0);
244244

245-
/* Write the page. If archiving/streaming, XLOG it. */
245+
/*
246+
* Write the page and log it. It might seem that an immediate sync
247+
* would be sufficient to guarantee that the file exists on disk, but
248+
* recovery itself might remove it while replaying, for example, an
249+
* XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we
250+
* need this even when wal_level=minimal.
251+
*/
246252
PageSetChecksumInplace(metapage,BTREE_METAPAGE);
247253
smgrwrite(index->rd_smgr,INIT_FORKNUM,BTREE_METAPAGE,
248254
(char*)metapage, true);
249-
if (XLogIsNeeded())
250-
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
251-
BTREE_METAPAGE,metapage, false);
255+
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
256+
BTREE_METAPAGE,metapage, false);
252257

253258
/*
254259
* An immediate sync is required even if we xlog'd the page, because the

‎src/backend/access/spgist/spginsert.c‎

Lines changed: 13 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -161,33 +161,36 @@ spgbuildempty(Relation index)
161161
page= (Page)palloc(BLCKSZ);
162162
SpGistInitMetapage(page);
163163

164-
/* Write the page. If archiving/streaming, XLOG it. */
164+
/*
165+
* Write the page and log it unconditionally. This is important
166+
* particularly for indexes created on tablespaces and databases
167+
* whose creation happened after the last redo pointer as recovery
168+
* removes any of their existing content when the corresponding
169+
* create records are replayed.
170+
*/
165171
PageSetChecksumInplace(page,SPGIST_METAPAGE_BLKNO);
166172
smgrwrite(index->rd_smgr,INIT_FORKNUM,SPGIST_METAPAGE_BLKNO,
167173
(char*)page, true);
168-
if (XLogIsNeeded())
169-
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
170-
SPGIST_METAPAGE_BLKNO,page, false);
174+
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
175+
SPGIST_METAPAGE_BLKNO,page, false);
171176

172177
/* Likewise for the root page. */
173178
SpGistInitPage(page,SPGIST_LEAF);
174179

175180
PageSetChecksumInplace(page,SPGIST_ROOT_BLKNO);
176181
smgrwrite(index->rd_smgr,INIT_FORKNUM,SPGIST_ROOT_BLKNO,
177182
(char*)page, true);
178-
if (XLogIsNeeded())
179-
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
180-
SPGIST_ROOT_BLKNO,page, true);
183+
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
184+
SPGIST_ROOT_BLKNO,page, true);
181185

182186
/* Likewise for the null-tuples root page. */
183187
SpGistInitPage(page,SPGIST_LEAF |SPGIST_NULLS);
184188

185189
PageSetChecksumInplace(page,SPGIST_NULL_BLKNO);
186190
smgrwrite(index->rd_smgr,INIT_FORKNUM,SPGIST_NULL_BLKNO,
187191
(char*)page, true);
188-
if (XLogIsNeeded())
189-
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
190-
SPGIST_NULL_BLKNO,page, true);
192+
log_newpage(&index->rd_smgr->smgr_rnode.node,INIT_FORKNUM,
193+
SPGIST_NULL_BLKNO,page, true);
191194

192195
/*
193196
* An immediate sync is required even if we xlog'd the pages, because the

‎src/backend/catalog/heap.c‎

Lines changed: 7 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1370,18 +1370,19 @@ heap_create_with_catalog(const char *relname,
13701370

13711371
/*
13721372
* Set up an init fork for an unlogged table so that it can be correctly
1373-
* reinitialized on restart. Since we're going to do an immediate sync, we
1374-
* only need to xlog this if archiving or streaming is enabled. And the
1375-
* immediate sync is required, because otherwise there's no guarantee that
1376-
* this will hit the disk before the next checkpoint moves the redo pointer.
1373+
* reinitialized on restart. An immediate sync is required even if the
1374+
* page has been logged, because the write did not go through
1375+
* shared_buffers and therefore a concurrent checkpoint may have moved
1376+
* the redo pointer past our xlog record. Recovery may as well remove it
1377+
* while replaying, for example, XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE
1378+
* record. Therefore, logging is necessary even if wal_level=minimal.
13771379
*/
13781380
void
13791381
heap_create_init_fork(Relationrel)
13801382
{
13811383
RelationOpenSmgr(rel);
13821384
smgrcreate(rel->rd_smgr,INIT_FORKNUM, false);
1383-
if (XLogIsNeeded())
1384-
log_smgrcreate(&rel->rd_smgr->smgr_rnode.node,INIT_FORKNUM);
1385+
log_smgrcreate(&rel->rd_smgr->smgr_rnode.node,INIT_FORKNUM);
13851386
smgrimmedsync(rel->rd_smgr,INIT_FORKNUM);
13861387
}
13871388

0 commit comments

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