On Fri, Nov 2, 2018 at 7:37 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> Robert Haas <robertmhaas@gmail.com> writes:
> > That's not what I'm saying. If we don't have the FSM, we have to
> > check every page of the table. If there's a workload where that
> > happens a lot on a table that is just under the size threshold for
> > creating the FSM, then it's likely to be a worst case for this patch.
>
> Hmm, you're assuming something not in evidence: why would that be the
> algorithm? On a FSM-less table, I'd be inclined to just check the
> last page and then grow the table if the tuple doesn't fit there.
> This would, in many cases, soon result in a FSM being created, but
> I think that's just fine. The point of the change is to optimize
> for cases where a table *never* gets more than a few inserts. Not, IMO,
> for cases where a table gets a lot of churn but never has a whole lot of
> live tuples. In the latter scenario we are far better off having a FSM.
>
In the past, you seem to have suggested an approach to try each block
[1] for small tables which don't have FSM. I think if we do what you
are suggesting now, then we don't need to worry much about any
regression and code will be somewhat simpler, but OTOH, I don't see
much harm in trying every block if we keep the threshold as no more
than 4. That would address more cases.
[1] - https://www.postgresql.org/message-id/11360.1345502641%40sss.pgh.pa.us
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com