Re: Hot standby, RestoreBkpBlocks and cleanup locks

Simon Riggs wrote:
> The idea outlined before didn't deal with all call points for
> RecordIsCleanupRecord(), so doesn't actually work.

Hmm, grep finds two call points for that:

> ! case RECOVERY_TARGET_PAUSE_CLEANUP:
> ! /*
> ! * Advance until we see a cleanup record, then pause.
> ! */
> ! if (RecordIsCleanupRecord(record))
> ! pauseHere = true;
> ! break;
> !

and

> + /*
> + * Wait, kill or otherwise resolve any conflicts between
> + * incoming cleanup records and user queries. This is the
> + * main barrier that allows MVCC to work correctly when
> + * running standby servers. Only need to do this if there
> + * is a possibility that users may be active.
> + */
> + if (reachedSafeStartPoint && RecordIsCleanupRecord(record))
> + ResolveRedoVisibilityConflicts(EndRecPtr, record);

The second we can easily handle by getting rid of
ResolveRedoVisibilityConflicts functions and making the redo-functions
call XactResolveRecoveryConflicts when necessary.

Is the first really useful? I would understand "advance until next
cleanup record *that would block/kill queries*", but why would you want
to advance until the next cleanup record? Anyway, if it is useful, you
could do the pausing in XactResolveRecoveryConflicts, too.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com