Re: Vacuum full on a big table

Gaetano Mendola <mendola(at)bigfoot(dot)com> writes:
> Scott Marlowe wrote:
>> Is there a reason you're doing a full vacuum?

> Because I'm only running pg_autovacuum since one month now, but I see
> that for same table is a disaster do not vacuum full once in a day.

You need to find out why regular vacuum isn't getting the job done,
not look for a bigger hammer. What is growing, exactly --- the table,
its indexes (all of them, or only some), the TOAST table or index?
Do you have adequate FSM space according to what VACUUM VERBOSE says?

It's much more likely that we'd remove VACUUM FULL entirely than do
major work on it (like try to make it work on part of a table, which
is something I don't believe could work anyway). That entire approach
is fundamentally broken when it comes to major rearrangements of huge
tables: it requires exclusive lock, it takes forever, and it bloats the
table's indexes.

regards, tom lane