On Thu, May 27, 2010 at 2:17 PM, Josh Berkus <josh@agliodbs.com> wrote:
> On 5/26/10 6:32 PM, Robert Haas wrote:
>> Hmm, yeah. Maybe we should freeze when we set PD_ALL_VISIBLE; that
>> might be just as good, and simpler. Assuming the visibility map is
>> sufficiently crash-safe/non-buggy, we could then teach VACUUM that
>> it's OK to advance relfrozenxid even when doing just a partial vacuum
>> - because any pages that were skipped must contain only frozen tuples.
>> Previously you've objected to proposals in this direction because
>> they might destroy forensic information, but maybe we should do it
>> anyway.
>
> It would be an improvement, and easier than the various ways of never
> having to visit the pages, which are all fairly intensive. Given the
> destruction of rollback information, though, we'd probably want a way to
> switch this behaviour on and off as an autovacuum setting.
It's not going to destroy anything that is needed for rollback unless
there's a bug - PD_ALL_VISIBLE only gets set when all tuples on the
page are visible to all backends. That can't happen until all
transactions that wrote the page, and all others that have a lower
xmin, have committed. That having been said, if making it a GUC makes
people less nervous about doing it, then +1 from me.
> Does this send us down the wrong path, though? I thought we wanted to
> think about removing hint bits so that we could implement things like
> CRCs. No?
PD_ALL_VISIBLE is a page-level bit, not a tuple-level bit, and I
strongly suspect it's not going anywhere. It's critical
infrastructure for index-only scans, among other things.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company