On Thu, 2009-09-24 at 13:33 +0300, Heikki Linnakangas wrote:
> Heikki Linnakangas wrote:
> > The problem becomes a lot easier if we accept that it's OK to have a
> > lock included in the running-xacts snapshot and also appear in a
> > XLOG_RELATION_LOCK record later. The standby should handle that
> > gracefully already. If we just remove RecoveryInfoLock, that can happen,
> > but it still won't be possible for a lock to be missed out which is what
> > we really care about.
>
> I see the problem with that now. Without the lock, it's possible that
> the XLOG_RELATION_LOCK WAL record is written before the
> XLOG_RUNNING_XACTS record. If the lock is not included in the snapshot,
> we want the lock WAL record to be after the snapshot record.
>
> So i guess we'll need the RecoveryInfoLock. But we don't need to hold it
> across the wait. I think it's enough to acquire it just before writing
> the WAL record in LockAcquire. That will ensure that the WAL record
> isn't written until the snapshot is completely finished.
I will think some more on that. I remember thinking there was a reason
why that wasn't enough, possibly to do with no-wait locks which I
remember caused me to change that code.
-- Simon Riggs www.2ndQuadrant.com
