Shaun Thomas wrote:
> I don't notice any errors, which just makes this even more strange.
> But after weeks of operating normally, our 10pm manual vacuum job
> generated transaction logs basically equivalent to 3/4 of our
> database, and I can't find any explanation. This amount is about 6x
> higher than usual.
>
> Before I go crazy and tear the box apart, does anyone know of some
> internal change that may cause intermittent marking of pages to
> increase without related database activity?
Yes.
Did you bulk load this data (possibly through restoring pg_dump
output)? If so, and you have not explicitly run VACUUM FREEZE
afterward, the vacuum noticed that it was time to freeze all of these
tuples. When I use pg_dump output to create a database, I always
VACUUM FREEZE ANALYZE immediately afterward -- before I start
archiving.
Just be glad you got it with a manual vacuum during off-peak hours,
rather than having this kick in via autovacuum during peak OLTP load.
> Or did I just vacuum a database with a corrupt CPU or piece of RAM?
> (Let me say again, I see no errors anywhere in the database logs.)
You haven't mentioned anything that should be taken as evidence of
corruption or any unusual behavior on the part of PostgreSQL.
-Kevin