"Josh Berkus" <josh@agliodbs.com> writes:
> Damn. I can only really help with the obvious things, and you've
> covered most of those. That view should take a long time, given the
> complexity ... but a long time is 30-60 seconds, not 10 minutes.
I'm guessing it's ye olde stupid-choice-of-plan kind of problem.
Unless I missed it, we have not been shown the plan currently being
used (no, the first four lines don't do it for me...)
You could probably get back to the previous, not-so-slow plan choice
by doing "DELETE FROM pg_statistic" to remove the data collected by
VACUUM ANALYZE. A comparison of the EXPLAIN output in that state
vs. the EXPLAIN output in the analyzed state might be enlightening.
regards, tom lane