On Tue, Jun 2, 2015 at 11:44 AM, Andres Freund <andres@anarazel.de> wrote:
> On 2015-06-02 11:37:02 -0400, Robert Haas wrote:
>> The exact circumstances under which we're willing to replace a
>> relminmxid with a newly-computed one that differs are not altogether
>> clear to me, but there's an "if" statement protecting that logic, so
>> there are some circumstances in which we'll leave the existing value
>> intact.
>
> I guess we'd have to change that then.
Yeah, but first we'd need to assess why it's like that. Tom was the
one who installed the current logic, but I haven't been able to fully
understand it.
>> It would similarly do so when the oldest MXID reference in the
>> relation is in fact 1, but that value can't be vacuumed away yet.
>
> I'd thought of just forcing consumption of one multixact in that
> case. Not pretty, but imo acceptable.
What if multixact 1 still has living members?
>> Also, the database might be really big. Even if it were true that a
>> full scan of every table would get us out of this state, describing
>> the time that it would take to do that as "relatively short" seems to
>> me to be considerably understating the impact of a full-cluster
>> VACUUM.
>
> Well. We're dealing with a corrupted cluster. Having a way out that's
> done automatically, even if it takes a long while, is pretty good
> already. In many cases the corruption (i.e. pg_upgrade) happened long
> ago, so the table's relminmxid will already have been recomputed. I
> think that's acceptable.
I'm a long way from being convinced the automated recovery is
possible. There are so many different scenarios here that it's very
difficult to reason generally about what the "right" thing to do is.
I agree it would be nice if we had it, though.
>> With regard to the more general question of WAL-logging this, are you
>> going to work on that? Are you hoping Alvaro or I will work on that?
>> Should we draw straws? It seems like somebody needs to do it.
>
> I'm willing to invest the time to develop an initial version, but I'll
> need help evaluating it. I don't have many testing resources available
> atm, and I'm not going to trust stuff I developed while travelling by
> just looking at the code.
I'm willing to help with that. Hopefully I'm not the only one, though.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company