Jochen Erwied <jochen@pgsql.erwied.eu> writes:
> Monday, October 25, 2010, 4:12:39 PM you wrote:
>> "Jochen Erwied" <jochen@pgsql.erwied.eu> writes:
>>> VACUUM FULL does not update statistics so display of pg_stat_user_tables is
>>> wrong. A normal VACUUM updates the relevant information.
>> Hmm. This is a definitional issue: what do we really mean by last_vacuum?
>> I'm inclined to think that the current behavior is reasonable. VACUUM
>> FULL is (still) not intended as a routine maintenance operation, and
>> the point of that column is to track routine maintenance operations.
> Well, when reading
> http://www.postgresql.org/docs/current/static/monitoring-stats.html
> then last_vacuum contains the last time of a user-initiated vacuum. There's
> no distinction made what kind of vacuum was made. And IMHO even if VACUUM
> FULL isn't meant for routine vacuuming, the state should be changed.
Perhaps. The new implementation of VACUUM FULL is really more like a
CLUSTER, or one of the rewriting variants of ALTER TABLE. Should all
of those operations result in an update of last_vacuum? From an
implementation standpoint it's difficult to say that only some of them
should, because all of them result in a table that has no immediate
need for vacuuming. The only argument I can see for having only VACUUM
FULL update the timestamp is that it's called VACUUM and the others
aren't. Which is an argument, but not a terribly impressive one IMO.
> Of course the easiest way to fix this bug (or better flaw) is to change the
> documentation :-)
Yeah, that part of the docs will require editing no matter what we do.
I'm just trying to get some clarity on what the most reasonable behavior
is.
regards, tom lane