Re: VACUUM ANALYSE...

"Thilo Hille" <thilo(at)resourcery(dot)de> writes:
> shared_buffers =70000
> max_fsm_relations = 100
> max_fsm_pages = 2000
> sort_mem = 128
> vacuum_mem = 8192

If your DB is large enough that it takes an hour to run VACUUM, then
those FSM parameters are surely way too small. I'd try something
like 1000/1000000 for starters.

Also, boosting vacuum_mem might help speed up VACUUM, if you have a
reasonable amount of RAM in the box. (Instead of 8192 = 8Mb, try
50000 or so.)

BTW, what *is* the amount of RAM in the box? I'm eyeing the
shared_buffers setting with suspicion. It may be too high.
500Mb in shared buffers would very likely be more usefully spent
elsewhere.

It's very likely that the undersized FSM settings have caused the system
to leak a lot of disk space, and that the only way to recover it will
now be a VACUUM FULL. Which will be painful :-(. Can you show us the
output of VACUUM VERBOSE for your larger tables?

regards, tom lane