Re: Database OID xxxxx now seems to belong to "foo"

Tom Lane wrote:
> Richard Huxton <dev(at)archonet(dot)com> writes:
>> Gauthier, Dave wrote:
>>>> Trying (and failing) to attach to my DBs. Getting...
>>>> database "foo_standby" has disappeared form pg_database
>>>> DETAIL: Database OID 2323523 now seems to belong to "foo"
>
>> Hmm - if a shutdown + restart fixed it, I'm wondering if it wasn't just
>> a long-lived connection remembering where 2323523 used to point to.
>
> No, it's the "flat file" copy of pg_database that's supplying that
> number, and the reason the restart fixed it is that the flat file
> is forcibly rebuilt during a restart. What's not quite clear is
> why the flat file was wrong.

Ah, that makes sense (well, the first part).

> We've seen this type of failure reported from the field before,
> and as far as I recall the triggering condition was transaction ID
> wraparound due to lack of vacuuming ... but haven't consumed enough
> caffeine this morning to remember details.

Be interesting to find out - I can't quite imagine how a transaction ID
problem would cause this.

--
Richard Huxton
Archonet Ltd