Re: Reaping Temp tables to avoid XID wraparound

On Mon, Feb 18, 2019 at 2:31 AM Michael Paquier <michael(at)paquier(dot)xyz> wrote:

> On Sun, Feb 17, 2019 at 05:47:09PM +0100, Magnus Hagander wrote:
> > We could I guess add a field specifically for temp_namespace_xid or such.
> > The question is if it's worth the overhead to do that.
>
> That would mean an extra 4 bytes in PGPROC, which is something we
> could live with, still the use-case looks rather narrow to me to
> justify that.
>

It does, tha'ts why I questioned if it's worth it. But, thinking some more
about it, some other options would be:

1. This is only set once per backend in normal operations, right? (Unless I
go drop the schema manually, but that's not exactly normal). So maybe we
could invent a pg stat message and send the information through the
collector? Since it doesn't have to be frequently updated, like your
typical backend_xmin.

2. Or probably even better, just put it in PgBackendStatus? Overhead here
is a lot cheaper than PGPROC.

ISTM 2 is probably the most reasonable option here?

--
Magnus Hagander
Me: https://www.hagander.net/ <http://www.hagander.net/>
Work: https://www.redpill-linpro.com/ <http://www.redpill-linpro.com/>