Re: Plans for solving the VACUUM problem

mlw <markw(at)mohawksoft(dot)com> writes:
> My only suggestion would be to store some information in the statistics about
> whether or not, and how bad, a table needs to be vacuumed.

I was toying with the notion of using the FSM to derive that info,
somewhat indirectly to be sure (since what the FSM could tell you would
be about tuples inserted not tuples deleted). Heavily used FSM entries
would be the vacuum daemon's cues for tables to hit more often.

ANALYZE stats don't seem like a productive way to attack this, since
there's no guarantee they'd be updated often enough. If the overall
data distribution of a table isn't changing much, there's no need to
analyze it often.

> Also, it would be good to be able to update the statistics without doing a
> vacuum, i.e. rather than having to vacuum to analyze, being able to analyze
> without a vacuum.

Irrelevant, not to mention already done ...

regards, tom lane