database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-06-29 10:41:53 +03:00
Code Activity

Fix failure to guarantee that a checkpoint will write out pg_clog updates

Browse Source 
for transaction commits that occurred just before the checkpoint.  This is
an EXTREMELY serious bug --- kudos to Satoshi Okada for creating a
reproducible test case to prove its existence.
This commit is contained in:
Tom Lane
2004-08-11 04:07:16 +00:00
parent bc8a1fc282
commit 3fdf649f4f
3 changed files with 45 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
src/backend/access/transam/xlog.c
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
* $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.158 2004/08/09 16:26:01 tgl Exp $
* $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.159 2004/08/11 04:07:15 tgl Exp $
*
*-------------------------------------------------------------------------
*/
@ -4699,6 +4699,15 @@ CreateCheckPoint(bool shutdown, bool force)
checkPoint.ThisTimeLineID = ThisTimeLineID;
checkPoint.time = time(NULL);
/*
* We must hold CheckpointStartLock while determining the checkpoint
* REDO pointer. This ensures that any concurrent transaction commits
* will be either not yet logged, or logged and recorded in pg_clog.
* See notes in RecordTransactionCommit().
*/
LWLockAcquire(CheckpointStartLock, LW_EXCLUSIVE);
/* And we need WALInsertLock too */
LWLockAcquire(WALInsertLock, LW_EXCLUSIVE);
/*
@ -4731,6 +4740,7 @@ CreateCheckPoint(bool shutdown, bool force)
ControlFile->checkPointCopy.redo.xrecoff)
{
LWLockRelease(WALInsertLock);
LWLockRelease(CheckpointStartLock);
LWLockRelease(CheckpointLock);
END_CRIT_SECTION();
return;
@ -4789,6 +4799,9 @@ CreateCheckPoint(bool shutdown, bool force)
* GetSnapshotData needs to get XidGenLock while holding SInvalLock,
* so there's a risk of deadlock. Need to find a better solution. See
* pgsql-hackers discussion of 17-Dec-01.
*
* XXX actually, the whole UNDO code is dead code and unlikely to ever
* be revived, so the lack of a good solution here is not troubling.
*/
#ifdef NOT_USED
checkPoint.undo = GetUndoRecPtr();
@ -4798,11 +4811,13 @@ CreateCheckPoint(bool shutdown, bool force)
#endif
/*
* Now we can release insert lock, allowing other xacts to proceed
* even while we are flushing disk buffers.
* Now we can release insert lock and checkpoint start lock, allowing
* other xacts to proceed even while we are flushing disk buffers.
*/
LWLockRelease(WALInsertLock);
LWLockRelease(CheckpointStartLock);
/*
* Get the other info we need for the checkpoint record.
*/