Re: Visibility map thoughts

On 11/6/07, Heikki Linnakangas <heikki(at)enterprisedb(dot)com> wrote:
> Marko Kreen wrote:
> > On 11/6/07, Heikki Linnakangas <heikki(at)enterprisedb(dot)com> wrote:
> >> (Gosh, we really need a name for the sort of vacuum. I was about to say
> >> "we'd still need regular regular VACUUMs" :-))
> >
> > As the new VACUUM variant will be somewhat unsafe, it should
> > not replace "regular" VACUUM but get separate name.
>
> What do you mean by unsafe? It is supposed to reclaim all dead tuples a
> normal vacuum would, except for HOT updated tuples that can be pruned
> without scanning indexes. It doesn't advance the relfrozenxid or update
> stats, though.

I got the impression from this paragraph:

"This kind of map is useful for VACUUM as well, though VACUUM could get
away with less strict semantics, and we might not want to visit a page
in VACUUM if there's only very few dead tuples on a page."

That seems to hint that the current VACUUM needs still be
run occasionally.

> > VACUUM FAST maybe? Informally "fastvacuum". Something with
> > "lazy" or "partial" would also be possibility.
>
> We already call the regular vacuum "lazy" in the source code, as opposed
> to VACUUM FULL. Partial is also bit misleading; while it doesn't scan
> the whole table, it should find all dead tuples.

As the VACUUM FULL is unusable in real life, maybe we should
move current VACUUM functionality under VACUUM FULL ;)

--
marko