Re: Partial vacuum versus pg_class.reltuples

On Sat, Jun 6, 2009 at 3:44 PM, Tom Lane<tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> I complained a couple days ago that in HEAD, vacuum is putting
> very bogus values into pg_class.reltuples for indexes:
> http://archives.postgresql.org/pgsql-bugs/2009-06/msg00037.php
>
> After looking through the code a bit, I've confirmed my prior guess that
> this is caused by the partial-vacuum patch.  The number that's being
> used is the number of live tuples found on the pages that were actually
> scanned.  There's a kluge in there to keep from storing this number as
> the relation's own reltuples, but the implications for indexes were not
> considered.  The index AMs generally assume that what they're told in
> IndexVacuumInfo.num_heap_tuples can be trusted, and that's not true
> anymore.
>
> I think probably the only workable solution for 8.4 is to use the prior
> value of the relation's reltuples field as num_heap_tuples when we have
> not scanned the whole heap.  This will effectively mean that index
> reltuples values don't change either in a partial vacuum.

Am I wrong to be frightened by the implications of updating this value
only once in a blue moon? Doesn't this have the potential to result
in really bad plans? Do we have any reasonable manual way of forcing
VACUUM to scan the entire heap?

...Robert