DW <dwinner-lists@att.net> writes:
> I'm perplexed. I'm trying to find out why some queries are taking a long
> time, and have found that after running analyze, one particular query
> becomes slow.
This implies that the planner's default choice of plan (without any
statistics) is better than its choice when informed by statistics.
This is undesirable but not unheard of :-(
It would be interesting to see EXPLAIN ANALYZE results in both cases,
plus the contents of the relevant pg_stats rows. (BTW, you need not
dump and reload to get back to the virgin state --- just delete the
relevant rows from pg_statistic.) Also we'd want to know exactly what
PG version this is, and on what sort of platform.
You might be able to fix things by increasing the statistics targets or
tweaking planner cost parameters, but it'd be best to investigate before
trying to fix.
regards, tom lane