Re: Missing CONCURRENT VACUUM (Was: Release notes for

On E, 2005-08-22 at 10:03 -0400, Tom Lane wrote:
> Hannu Krosing <hannu(at)tm(dot)ee> writes:
> > Please try to apply the patch, even if dangling BEGIN WORK; still causes
> > problems - I'll fix this tomorrow.
>
> No. A patch that you yourself have so little confidence in, in a
> fundamental part of the system?

Yeah, I know I sound flaky. Never should I submit patches when
terminally tired :( . I had just recently read something about starting
beta on monday and wanted to get it in before that.

I had quite much confidence in it not breaking anything (at least not
anything more than what the previous crippled patch did), I had just to
fix the parts regarding initialising nonInVacuumXmin, which is done in
todays patch.

> This will be lucky if it gets into 8.2, after considerably more review
> than I have time to give it now. It clearly needs such review, and we
> are way too late with the beta already.

When I started with the patch a long time ago, GetOldestXmin was used in
less places than it is now. Since then more functionality seems to have
started using it without paying attention that it is used for two
logically different things.

I hope that some care will be taken in future to distinguish between
"real" transactions (those that modify data and whose actions can be
rollbacked) and vacuum (and maybe some other maintenance tasks) where
transaction is used as a programming convenience without any real
transaction-like behaviour present.

Getting my todays patch in would be a way to make sure this distinction
will get more attention.

--
Hannu Krosing <hannu(at)tm(dot)ee>