On Wed, Mar 29, 2017 at 2:23 AM, Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> I was thinking that the status of this patch is still "Needs review"
> because I sent latest version patch[1].
I think you're right.
I took a look at this today. I think there is some problem with the
design of this patch. I originally proposed a threshold based on the
percentage of not-all-visible pages on the theory that we'd just skip
looking at the indexes altogether in that case. But that's not what
the patch does: it only avoids the index *cleanup*, not the index
*vacuum*. And the comments in btvacuumcleanup say this:
/* * If btbulkdelete was called, we need not do anything, just return the * stats from the latest btbulkdelete
call. If it wasn't called, we must * still do a pass over the index, to recycle any newly-recyclable pages * and
toobtain index statistics. * * Since we aren't going to actually delete any leaf items, there's no * need to
gothrough all the vacuum-cycle-ID pushups. */
So, if I'm reading this correctly, the only time this patch saves
substantial work - at least in the case of a btree index - is in the
case where there are no dead tuples at all. But if we only want to
avoid the work in that case, then a threshold based on the percentage
of all-visible pages seems like the wrong thing, because the other
stuff btvacuumcleanup() is doing doesn't have anything to do with the
number of all-visible pages.
I'm not quite sure what the right thing to do is here, but I'm
doubtful that this is it.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company