Re: gincostestimate

On Wed, Sep 8, 2010 at 1:02 AM, Teodor Sigaev <teodor(at)sigaev(dot)ru> wrote:
> Fixed, and slightly reworked to be more clear.
> Attached patch is based on your patch.

The patch will improve accuracy of plans using gin indexes.
It only adds block-level statistics information into the meta
pages in gin indexes. Data-level statistics are not collected
by the patch, and there are no changes in pg_statistic.

The stats page is updated only in VACUUM. ANALYZE doesn't update
the information at all. In addition, REINDEX, VACUUM FULL, and
CLUSTER reset the information to zero, but the reset is not preferable.
Is it possible to fill the statistic fields at bulk index-build?
No one wants to run VACUUM after VACUUM FULL to update the GIN stats.

We don't have any methods to dump the meta information at all.
They might be internal information, but some developers and
debuggers might want such kinds of tools. Contrib/pageinspect
might be a good location to have such function; it has bt_metap().

The patch can be applied cleanly, no compiler warnings, and it passed
all existing regression tests. There are no additional documentation
and regression tests -- I'm not sure whether we should have them.
If the patch is an internal improvement, docs are not needed.

--
Itagaki Takahiro