Re: Freezing is not WAL-logged

Heikki Linnakangas wrote:
> Alvaro Herrera wrote:
> >Tom Lane wrote:
> >>"Heikki Linnakangas" <heikki(at)enterprisedb(dot)com> writes:
> >>>I just noticed that freezing a tuple (in vacuumlazy.c) is not
> >>>WAL-logged.
> >>The theory is that this doesn't matter because the tuple is committed
> >>either way ... it's equivalent to a hint-bit update which we don'ton
> >>WAL-log either.
> >
> >Also it'd be hugely expensive to log each freeze operation. The
> >alternative would be to log a VACUUM FREEZE, but that has the potential
> >to cause enormous recovery runtime.
>
> Freezing isn't very common, and a single WAL record per page would be
> enough. I can write the patch.

Yeah, you are right, I was thinking that we would freeze a lot of tuples
a lot of the time, but this is obviously not true.

> It does have the potential to increase recovery times, but I don't think
> we can just cross our fingers and hope that no crash happens after
> freezing some tuples.

Agreed.

My idea was to move relminxid to a new pg_class_nt non-transactional
catalog in 8.3, but we certainly need to give a solution for the problem
in 8.2. I don't remember why I concluded that we were safe against this
problem, but you are right that we aren't.

--
Alvaro Herrera http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support