On Fri, Jan 27, 2023 at 3:04 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Thomas Munro <thomas.munro@gmail.com> writes:
> > On Fri, Jan 27, 2023 at 1:30 PM Michael Paquier <michael@paquier.xyz> wrote:
> >> My opinion would be to make this function more reliable, FWIW, even if
> >> that involves a performance impact when called in a close loop by
> >> forcing more WAL flushes to ensure its report durability and
> >> consistency.
>
> > Yeah, the other thread has a patch for that. But it would hurt some
> > workloads.
>
> I think we need to get the thing correct first and worry about
> performance later. What's wrong with simply making pg_xact_status
> write and flush a record of the XID's existence before returning it?
> Yeah, it will cost you if you use that function, but not if you don't.
It would be pg_current_xact_id() that would have to pay the cost of
the WAL flush, not pg_xact_status() itself, but yeah that's what the
patch does (with some optimisations). I guess one question is whether
there are any other reasonable real world uses of
pg_current_xact_id(), other than the original goal[1]. If not, then
at least you are penalising the right users, even though they probably
only actually call pg_current_status() in extremely rare circumstances
(if COMMIT hangs up). But I wouldn't be surprised if people have
found other reasons to be interested in xid observability, related to
distributed transactions and snapshots and suchlike. There is no
doubt that the current situation is unacceptable, though, so maybe we
really should just do it and make a faster one later. Anyone else
want to vote on this?
[1]
https://www.postgresql.org/message-id/flat/CAMsr%2BYHQiWNEi0daCTboS40T%2BV5s_%2Bdst3PYv_8v2wNVH%2BXx4g%40mail.gmail.com