Re: Temporary tables prevent autovacuum, leading to XID wraparound

On Mon, Aug 13, 2018 at 01:53:18PM -0300, Alvaro Herrera wrote:
> I do share Andres' concerns on the wording the comment. I would say
> something like
>
> /*
> * Reset the temporary namespace flag in MyProc. We assume this to be
> * an atomic assignment.
> *
> * Because this subtransaction is rolling back, the pg_namespace
> * row is not visible to anyone else anyway, but that doesn't matter:
> * it's not a problem if objects contained in this namespace are removed
> * concurrently.
> */

> The fact of assignment being atomic and the fact of the pg_namespace row
> being visible are separately important. You care about it being atomic
> because it means you must not have someone read "16" (0x10) when you
> were partway removing the value "65552" (0x10010), thus causing that
> someone removing namespace 16. And you care about the visibility of the
> pg_namespace row because of whether you're worried about a third party
> removing the tables from that namespace or not: since the subxact is
> aborting, you are not.

I was thinking about adding "Even if it is not atomic" or such at the
beginning of the paragraph, but at the end your phrasing sounds better
to me. So I have hacked up the attached, which also reworks the comment
in InitTempTableNamespace in the same spirit. Thoughts?
--
Michael