"Kevin Grittner" <[email protected]> writes:
> Tom Lane <[email protected]> wrote:
>> The approach contemplated in the comment, of assembling some stats
>> on-the-fly from the stats for individual child tables, doesn't
>> seem real practical from a planning-time standpoint.
> Can you give a thumbnail sketch of why that is?
Well, it would be expensive, and it's not even clear you can do it at
all (merging histograms with overlapping bins seems like a mess for
instance).
I think we have previously discussed the idea of generating and storing
ANALYZE stats for a whole inheritance tree, which'd solve the problem
nicely from the planner's standpoint. I'm a bit worried about the
locking implications, but if it took just a SELECT lock on the child
tables it probably wouldn't be too bad --- no worse than any other
SELECT on the inheritance tree. Another thing that's hard to figure out
is how autovacuum would know when to redo the stats. In a lot of common
situations, the inheritance parent table is empty and never changes, so
no autovac or autoanalyze would ever get launched against it.
regards, tom lane
--
Sent via pgsql-performance mailing list ([email protected])
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
