I did create and drop an index at some point while looking at this
issue. But I definitely reran both of the queries (and explains) after
the index was dropped, so I don't understand why there would be a
difference between the inner and left query plans. (which were run
back-to-back more than once) Anyways - I'll let you know if something
similar happens again.
Thanks,
Jeremy Haile
On Wed, 10 Jan 2007 14:22:35 -0500, "Tom Lane" <tgl@sss.pgh.pa.us> said:
> "Jeremy Haile" <jhaile@fastmail.fm> writes:
> > Another random idea - does PostgreSQL do any caching of query plans?
>
> Only if the client specifies it, either by PREPARE or the equivalent
> protocol-level message. I dunno what client software you were using,
> but I think few if any would PREPARE behind your back. Might be worth
> checking into though, if you've eliminated autovacuum.
>
> Actually there's another possibility --- did you create any indexes on
> the table in between? CREATE INDEX doesn't do a full stats update, but
> it does count the rows and update pg_class.reltuples. But it's hard to
> believe that'd have caused as big a rowcount shift as we see here ...
>
> regards, tom lane