Re: snapbuild woes

On 01/05/17 04:29, Craig Ringer wrote:
> On 1 May 2017 at 09:54, Petr Jelinek <petr(dot)jelinek(at)2ndquadrant(dot)com> wrote:
>
>> But, I still think we need to restart the tracking after new
>> xl_running_xacts. Reason for that is afaics any of the catalog snapshots
>> that we assigned to transactions at the end of SnapBuildCommitTxn might
>> be corrupted otherwise as they were built before we knew one of the
>> supposedly running txes was actually already committed and that
>> transaction might have done catalog changes.
>
> Due to the race where LogStandbySnapshot() collects running-xacts info
> while a concurrent xact commits, such that the xl_xact_commit appears
> before the xl_running_xacts, but the xl_running_xacts still has the
> commited xact listed as running, right? Because we update PGXACT only
> after we write the commit to WAL, so there's a window where an xact is
> committed in WAL but not shown as committed in shmem.
>

Yes that's what the patch at hand tries to fix, but Andres approached it
from too simplistic standpoint as we don't only care about the exported
snapshot but whatever catalog snapshots we made for the transactions we
already track, unless I am missing something.

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