Re: Grouped Index Tuples

Ron Mayer wrote:
> Jim C. Nasby wrote:
>> On usage, ISTM it would be better to turn on GIT only for a clustered
>> index and not the PK? I'm guessing your automatic case is intended for
>> SERIAL PKs, but maybe it would be better to just make that explicit.
>
> Not necessarily; since often (in my tables at least) the data for
> come columns has some local grouping of similar values even though
> it's not the clustered index.

Yes, there's a lot of cases like that.

My real goal is to make it cheap enough in the case where there is no
clustering, that we could just enable it on all indexes by default. At
the moment, it looks like it's indeed near-zero cost when the table is
in random order, but the CPU overhead is too great in many workloads to
have it always enabled. More autotuning logic would be needed, or a
significant reduction in overhead.

But as it is, you can always turn it on explicitly if you think it'd help.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com