Josh Berkus <josh@agliodbs.com> writes:
> Hmmm ... problem is, per my last e-mail, the bug is not reproducable off of
> this particular database instance -- if I copy it to my laptop, the bug goes
> away.
It's presumably dependent on the contents of pg_statistic and the
relpages/reltuples counts in pg_class for the tables involved.
You could likely reproduce it by migrating that data to your laptop.
It would take a little bit of hacking to get the pg_statistic data
in (adjusting starelid for instance) but I think it's doable.
Note that the planner control settings (eg effective_cache_size) might
also need to be copied.
regards, tom lane