Tobias Brox <tobixen(at)gmail(dot)com> wrote:
> Sorry for all the stupid questions ;-)
I'm with Mark -- I didn't see nay stupid questions there.
Where I would start, though, is by checking the level of bloat. One
long-running query under load, or one query which updates or deletes
a large number of rows, can put you into this state. If you find
serious bloat you may need to schedule a maintenance window for
aggressive work (like CLUSTER) to fix it.
Even before that, however, I would spend some time looking at the
patterns of I/O under `vmstat 1` or `iostat 1` to get a sense of
where the bottlenecks are. If you time-stamp the rows from vmstat
you can match them up against events in your log and periods of
slow response.
-Kevin