Re: BUG #6661: out-of-order XID insertion in KnownAssignedXids

On Wednesday, May 23, 2012 06:27:48 PM Andres Freund wrote:
> On Wednesday, May 23, 2012 12:30:31 PM Valentine Gogichashvili wrote:
> > > > CEST,,0,LOG,00000,"startup process (PID 23804) exited with exit code
> > > > 1",,,,,,,,,""
> > > > 2012-05-22 21:20:27.939 CEST,,,23793,,4fbbe69e.5cf1,2,,2012-05-22
> > > > 21:18:54 CEST,,0,LOG,00000,"terminating any other active server
> > > > processes",,,,,,,,,""
> > >
> > > Could you provide a log with log_min_messages=DEBUG4? This will be
> > > quite long though...
> >
> > Hm... I have the data directory, but not the WAL files any more. Forgot
> > to back them up :(
>
> Too bad, I think without more information we won't be able to fix this. I
> feel bad for it but I actually you run into it again ;)
>
> Here is what I found strange, should somebody ever come back to the issue:
> - the standby was in a consistent state
> - according to the error annotation we were processing a XLOG_RUNNING_XACTS
> - ProcArrayApplyRecoveryInfo with STANDBY_SNAPSHOT_READY only does
> ExpireOldKnownAssignedTransactionIds(running->oldestRunningXid);
> StandbyReleaseOldLocks(running->xcnt, running->xids);
> before short-circuiting. How comes we call KnownAssignedXidsAdd in that
> case?
Looking over the code again reachedConsistency doesn't imply
STANDBY_SNAPSHOT_READY.

Andres

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services