On Wed, May 25, 2011 at 5:57 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>
> Alternatively, it's possible that we'd be better off vacuuming the
> table more often (say, autovacuum_vacuum_scale_factor=0.10 or 0.08 or
> something) but only doing the index scans every once in a while when
> enough dead line pointers have accumulated.
Thats precisely the reason I suggested separating heap and index
vacuums instead of a tight integration as we have now. If we don't
spool the dead line pointers in a separate area though, we would need
to make sure that index vacuum runs through the heap first to collect
the dead line pointers and then remove the corresponding index
pointers. We would need to also take into consideration the
implications on visibility map for any such scheme to work correctly
and efficiently.
Thanks,
Pavan
--
Pavan Deolasee
EnterpriseDB http://www.enterprisedb.com
