Re: lastOverflowedXid does not handle transaction ID wraparound

On Wed, Oct 20, 2021 at 4:00 AM Andrey Borodin <x4mmm(at)yandex-team(dot)ru> wrote:

>
>
> > 17 окт. 2021 г., в 21:55, Dmitry Dolgov <9erthalion6(at)gmail(dot)com>
> написал(а):
> > I wonder what would be side
> > effects of clearing it when the snapshot is not suboverfloved anymore?
>
> I think we should just invalidate lastOverflowedXid on every
> XLOG_RUNNING_XACTS if subxid_overflow == false. I can't find a reason not
> to do so.
>
>
On a replica, I think it's possible for lastOverflowedXid to be set even if
subxid_overflow is false on the primary and secondary (
https://github.com/postgres/postgres/blob/dc899146dbf0e1d23fb24155a5155826ddce34c9/src/backend/storage/ipc/procarray.c#L1327).
I thought subxid_overflow only gets set if there are more than
PGPROC_MAX_CACHED_SUBXIDS (64) used in a given transaction.

Should the replica be invalidating lastOverflowedXid if subxcnt goes to
zero in XLOG_RUNNING_XACTS? But if there's an outstanding snapshot with an
xmin that precedes lastOverflowedXid we might violate MVCC if we invalidate
this, so I wonder if we also need to check the snapshot with the lowest
xmin?