Quoting "Scott Marlowe" <scott.marlowe@gmail.com>:
> Whether or not max relations is the root of the login hang problem,
> you likely have gotten a fair bit of bloat in your database if your
> setting was too low by a factor of 10 for so long. You may need to
> look at recovering lost space in bloated tables and / or indexes. If
> the bloat is real bad, look at dumping / restoring the database for a
> fresh start. It's a pain because it requires downtime, but it's often
> faster than anything else for a badly bloated database.
This occurred just after a dump/restore, so things were nice and lean.
I have a suspicion this is related to bgwriter_* and checkpoint I/O.
I've tweaked bgwriter_* and am awaiting an opportunity to test this
again (busy reindexing all) and will report back. Interestingly, this
behaviour didn't occur pre-partitioning. It seems the sheer number of
additional relations tipped things over -- suddenly many defaults are
just too low and I'm having to dig into arcane settings.
This thread seems to be related:
http://archives.postgresql.org//pgsql-admin/2008-10/msg00041.php
Cheers
Henry