Re: found xmin from before relfrozenxid on pg_catalog.pg_authid

Hi,

On 2018-06-19 08:25:33 -0500, Jeremy Finzel wrote:
> I have a question related to this - and specifically, preventing the error
> until we have a patch :).

The patch has been committed to postgres, although no release has been
made including it yet.


> We are encountering this error every few weeks on one very high
> transaction db, and have to restart to fix it.

> If I read you correctly, the cache may never be invalidated for these
> catalogs even if I manually VACUUM them?  I was thinking if I routinely run
> VACUUM FREEZE on these tables in every database I might avoid the
> issue.

Correct, that won't help. In fact, it'll quite possibly make it worse.


> But given the cause of the issue, would that just make no difference and I
> will still hit the error eventually?

Yes. You might be able to get away with just removing the
pg_internal.init files before and after an explicit VACUUM on shared
system tables.

Greetings,

Andres Freund