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

Commiteddbf39

Browse files
committed
Extend yesterday's patch so that the bgwriter is also told to forget
pending fsyncs during DROP DATABASE. Obviously necessary in hindsight :-(
1 parent530b10c commiteddbf39

File tree

4 files changed

+124
-55
lines changed

4 files changed

+124
-55
lines changed

‎src/backend/commands/dbcommands.c‎

Lines changed: 8 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,7 @@
1313
*
1414
*
1515
* IDENTIFICATION
16-
* $PostgreSQL: pgsql/src/backend/commands/dbcommands.c,v 1.189 2007/01/16 13:28:56 alvherre Exp $
16+
* $PostgreSQL: pgsql/src/backend/commands/dbcommands.c,v 1.190 2007/01/17 16:25:01 tgl Exp $
1717
*
1818
*-------------------------------------------------------------------------
1919
*/
@@ -40,6 +40,7 @@
4040
#include"postmaster/bgwriter.h"
4141
#include"storage/freespace.h"
4242
#include"storage/procarray.h"
43+
#include"storage/smgr.h"
4344
#include"utils/acl.h"
4445
#include"utils/builtins.h"
4546
#include"utils/flatfiles.h"
@@ -643,6 +644,12 @@ dropdb(const char *dbname, bool missing_ok)
643644
*/
644645
FreeSpaceMapForgetDatabase(db_id);
645646

647+
/*
648+
* Tell bgwriter to forget any pending fsync requests for files in the
649+
* database; else it'll fail at next checkpoint.
650+
*/
651+
ForgetDatabaseFsyncRequests(db_id);
652+
646653
/*
647654
* On Windows, force a checkpoint so that the bgwriter doesn't hold any
648655
* open files, which would cause rmdir() to fail.

‎src/backend/postmaster/bgwriter.c‎

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -37,7 +37,7 @@
3737
*
3838
*
3939
* IDENTIFICATION
40-
* $PostgreSQL: pgsql/src/backend/postmaster/bgwriter.c,v 1.35 2007/01/1700:17:20 tgl Exp $
40+
* $PostgreSQL: pgsql/src/backend/postmaster/bgwriter.c,v 1.36 2007/01/1716:25:01 tgl Exp $
4141
*
4242
*-------------------------------------------------------------------------
4343
*/
@@ -103,7 +103,7 @@
103103
typedefstruct
104104
{
105105
RelFileNodernode;
106-
BlockNumbersegno;/*InvalidBlockNumber means "revoke" */
106+
BlockNumbersegno;/*see md.c for special values */
107107
/* might add a real request-type field later; not needed yet */
108108
}BgWriterRequest;
109109

@@ -695,16 +695,16 @@ RequestCheckpoint(bool waitforit, bool warnontime)
695695
* ForwardFsyncRequest
696696
*Forward a file-fsync request from a backend to the bgwriter
697697
*
698-
* segno specifies which segment (not block!) of the relation needs to be
699-
* fsync'd. If segno == InvalidBlockNumber, the meaning is to revoke any
700-
* pending fsync requests for the entire relation (this message is sent
701-
* when the relation is about to be deleted).
702-
*
703698
* Whenever a backend is compelled to write directly to a relation
704699
* (which should be seldom, if the bgwriter is getting its job done),
705700
* the backend calls this routine to pass over knowledge that the relation
706701
* is dirty and must be fsync'd before next checkpoint.
707702
*
703+
* segno specifies which segment (not block!) of the relation needs to be
704+
* fsync'd. (Since the valid range is much less than BlockNumber, we can
705+
* use high values for special flags; that's all internal to md.c, which
706+
* see for details.)
707+
*
708708
* If we are unable to pass over the request (at present, this can happen
709709
* if the shared memory queue is full), we return false. That forces
710710
* the backend to do its own fsync. We hope that will be even more seldom.

‎src/backend/storage/smgr/md.c‎

Lines changed: 106 additions & 46 deletions
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 1.126 2007/01/1700:17:21 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/storage/smgr/md.c,v 1.127 2007/01/1716:25:01 tgl Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -31,6 +31,10 @@
3131
/* interval for calling AbsorbFsyncRequests in mdsync */
3232
#defineFSYNCS_PER_ABSORB10
3333

34+
/* special values for the segno arg to RememberFsyncRequest */
35+
#defineFORGET_RELATION_FSYNC(InvalidBlockNumber)
36+
#defineFORGET_DATABASE_FSYNC(InvalidBlockNumber-1)
37+
3438
/*
3539
* On Windows, we have to interpret EACCES as possibly meaning the same as
3640
* ENOENT, because if a file is unlinked-but-not-yet-gone on that platform,
@@ -258,30 +262,7 @@ mdunlink(RelFileNode rnode, bool isRedo)
258262
* We have to clean out any pending fsync requests for the doomed relation,
259263
* else the next mdsync() will fail.
260264
*/
261-
if (pendingOpsTable)
262-
{
263-
/* standalone backend or startup process: fsync state is local */
264-
RememberFsyncRequest(rnode,InvalidBlockNumber);
265-
}
266-
elseif (IsUnderPostmaster)
267-
{
268-
/*
269-
* Notify the bgwriter about it. If we fail to queue the revoke
270-
* message, we have to sleep and try again ... ugly, but hopefully
271-
* won't happen often.
272-
*
273-
* XXX should we CHECK_FOR_INTERRUPTS in this loop? Escaping with
274-
* an error would leave the no-longer-used file still present on
275-
* disk, which would be bad, so I'm inclined to assume that the
276-
* bgwriter will always empty the queue soon.
277-
*/
278-
while (!ForwardFsyncRequest(rnode,InvalidBlockNumber))
279-
pg_usleep(10000L);/* 10 msec seems a good number */
280-
/*
281-
* Note we don't wait for the bgwriter to actually absorb the
282-
* revoke message; see mdsync() for the implications.
283-
*/
284-
}
265+
ForgetRelationFsyncRequests(rnode);
285266

286267
path=relpath(rnode);
287268

@@ -894,7 +875,8 @@ mdsync(void)
894875
* what we will do is retry the whole process after absorbing fsync
895876
* request messages again. Since mdunlink() queues a "revoke" message
896877
* before actually unlinking, the fsync request is guaranteed to be gone
897-
* the second time if it really was this case.
878+
* the second time if it really was this case. DROP DATABASE likewise
879+
* has to tell us to forget fsync requests before it starts deletions.
898880
*/
899881
do {
900882
HASH_SEQ_STATUShstat;
@@ -1043,17 +1025,58 @@ register_dirty_segment(SMgrRelation reln, MdfdVec *seg)
10431025
* We stuff the fsync request into the local hash table for execution
10441026
* during the bgwriter's next checkpoint.
10451027
*
1046-
* segno == InvalidBlockNumber is a "revoke" request: remove any pending
1047-
* fsync requests for the whole relation.
1028+
* The range of possible segment numbers is way less than the range of
1029+
* BlockNumber, so we can reserve high values of segno for special purposes.
1030+
* We define two: FORGET_RELATION_FSYNC means to drop pending fsyncs for
1031+
* a relation, and FORGET_DATABASE_FSYNC means to drop pending fsyncs for
1032+
* a whole database. (These are a tad slow because the hash table has to be
1033+
* searched linearly, but it doesn't seem worth rethinking the table structure
1034+
* for them.)
10481035
*/
10491036
void
10501037
RememberFsyncRequest(RelFileNodernode,BlockNumbersegno)
10511038
{
10521039
Assert(pendingOpsTable);
10531040

1054-
if (segno!=InvalidBlockNumber)
1041+
if (segno==FORGET_RELATION_FSYNC)
1042+
{
1043+
/* Remove any pending requests for the entire relation */
1044+
HASH_SEQ_STATUShstat;
1045+
PendingOperationEntry*entry;
1046+
1047+
hash_seq_init(&hstat,pendingOpsTable);
1048+
while ((entry= (PendingOperationEntry*)hash_seq_search(&hstat))!=NULL)
1049+
{
1050+
if (RelFileNodeEquals(entry->tag.rnode,rnode))
1051+
{
1052+
/* Okay, delete this entry */
1053+
if (hash_search(pendingOpsTable,&entry->tag,
1054+
HASH_REMOVE,NULL)==NULL)
1055+
elog(ERROR,"pendingOpsTable corrupted");
1056+
}
1057+
}
1058+
}
1059+
elseif (segno==FORGET_DATABASE_FSYNC)
1060+
{
1061+
/* Remove any pending requests for the entire database */
1062+
HASH_SEQ_STATUShstat;
1063+
PendingOperationEntry*entry;
1064+
1065+
hash_seq_init(&hstat,pendingOpsTable);
1066+
while ((entry= (PendingOperationEntry*)hash_seq_search(&hstat))!=NULL)
1067+
{
1068+
if (entry->tag.rnode.dbNode==rnode.dbNode)
1069+
{
1070+
/* Okay, delete this entry */
1071+
if (hash_search(pendingOpsTable,&entry->tag,
1072+
HASH_REMOVE,NULL)==NULL)
1073+
elog(ERROR,"pendingOpsTable corrupted");
1074+
}
1075+
}
1076+
}
1077+
else
10551078
{
1056-
/*Enter a request to fsync this segment */
1079+
/*Normal case: enter a request to fsync this segment */
10571080
PendingOperationTagkey;
10581081
PendingOperationEntry*entry;
10591082
boolfound;
@@ -1070,29 +1093,66 @@ RememberFsyncRequest(RelFileNode rnode, BlockNumber segno)
10701093
if (!found)/* new entry, so initialize it */
10711094
entry->failures=0;
10721095
}
1073-
else
1096+
}
1097+
1098+
/*
1099+
* ForgetRelationFsyncRequests -- ensure any fsyncs for a rel are forgotten
1100+
*/
1101+
void
1102+
ForgetRelationFsyncRequests(RelFileNodernode)
1103+
{
1104+
if (pendingOpsTable)
1105+
{
1106+
/* standalone backend or startup process: fsync state is local */
1107+
RememberFsyncRequest(rnode,FORGET_RELATION_FSYNC);
1108+
}
1109+
elseif (IsUnderPostmaster)
10741110
{
10751111
/*
1076-
* Remove any pending requests for the entire relation. (This is a
1077-
* tad slow but it doesn't seem worth rethinking the table structure.)
1112+
* Notify the bgwriter about it. If we fail to queue the revoke
1113+
* message, we have to sleep and try again ... ugly, but hopefully
1114+
* won't happen often.
1115+
*
1116+
* XXX should we CHECK_FOR_INTERRUPTS in this loop? Escaping with
1117+
* an error would leave the no-longer-used file still present on
1118+
* disk, which would be bad, so I'm inclined to assume that the
1119+
* bgwriter will always empty the queue soon.
10781120
*/
1079-
HASH_SEQ_STATUShstat;
1080-
PendingOperationEntry*entry;
1121+
while (!ForwardFsyncRequest(rnode,FORGET_RELATION_FSYNC))
1122+
pg_usleep(10000L);/* 10 msec seems a good number */
1123+
/*
1124+
* Note we don't wait for the bgwriter to actually absorb the
1125+
* revoke message; see mdsync() for the implications.
1126+
*/
1127+
}
1128+
}
10811129

1082-
hash_seq_init(&hstat,pendingOpsTable);
1083-
while ((entry= (PendingOperationEntry*)hash_seq_search(&hstat))!=NULL)
1084-
{
1085-
if (RelFileNodeEquals(entry->tag.rnode,rnode))
1086-
{
1087-
/* Okay, delete this entry */
1088-
if (hash_search(pendingOpsTable,&entry->tag,
1089-
HASH_REMOVE,NULL)==NULL)
1090-
elog(ERROR,"pendingOpsTable corrupted");
1091-
}
1092-
}
1130+
/*
1131+
* ForgetDatabaseFsyncRequests -- ensure any fsyncs for a DB are forgotten
1132+
*/
1133+
void
1134+
ForgetDatabaseFsyncRequests(Oiddbid)
1135+
{
1136+
RelFileNodernode;
1137+
1138+
rnode.dbNode=dbid;
1139+
rnode.spcNode=0;
1140+
rnode.relNode=0;
1141+
1142+
if (pendingOpsTable)
1143+
{
1144+
/* standalone backend or startup process: fsync state is local */
1145+
RememberFsyncRequest(rnode,FORGET_DATABASE_FSYNC);
1146+
}
1147+
elseif (IsUnderPostmaster)
1148+
{
1149+
/* see notes in ForgetRelationFsyncRequests */
1150+
while (!ForwardFsyncRequest(rnode,FORGET_DATABASE_FSYNC))
1151+
pg_usleep(10000L);/* 10 msec seems a good number */
10931152
}
10941153
}
10951154

1155+
10961156
/*
10971157
*_fdvec_alloc() -- Make a MdfdVec object.
10981158
*/

‎src/include/storage/smgr.h‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
* Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
10-
* $PostgreSQL: pgsql/src/include/storage/smgr.h,v 1.57 2007/01/05 22:19:58 momjian Exp $
10+
* $PostgreSQL: pgsql/src/include/storage/smgr.h,v 1.58 2007/01/17 16:25:01 tgl Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -106,6 +106,8 @@ extern void mdimmedsync(SMgrRelation reln);
106106
externvoidmdsync(void);
107107

108108
externvoidRememberFsyncRequest(RelFileNodernode,BlockNumbersegno);
109+
externvoidForgetRelationFsyncRequests(RelFileNodernode);
110+
externvoidForgetDatabaseFsyncRequests(Oiddbid);
109111

110112
/* smgrtype.c */
111113
externDatumsmgrout(PG_FUNCTION_ARGS);

0 commit comments

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