Brian Fehrle <brianf@consistentstate.com> writes:
> Update on this:
> We did a switchover to another machine with the same hardware, however
> this system was running on some older parameters we had set in the
> postgresql.conf file.
> So we went from 400 max_connections to 200 max_connections, and 160MB
> work_mem to 200MB work_mem. And now on this other system, so far it
> seems to be running ok.
> Other than the obvious fact that each connection has a certain amount of
> memory usage, is there something else to watch for when increasing
> connections to numbers like 400? When we had the issue of the system
> jumping to 100% cpu usage, even at that point our number of backends to
> the cluster was at MAX 250, but generally in the 175 range, so well
> below our 400 max_connections we allow. So could this be the culprit?
Well, yeah, the pre-8.4 sinval problems I mentioned scale with the
number of live backends. When you have many more backends in the system
that will contribute to the problem, even --- in fact, especially --- if
the extra ones are idle.
regards, tom lane
