Tom Lane wrote:
Brian Hurt <bhurt@janestcapital.com> writes:
Is this a bug in postgres?
Well, if you'd provide enough info for someone else to reproduce it, we
could have a look.
Hello- I just wanted to close the book on this problem. I first noticed the problem when long copies kept slowing down. It turns out it wasn't a problem with Postgresql at all, as I recreated it with bonnie++ (which explains the long silence on this issue). After doing a lot of sustained I/O, we see I/O wait times climb until we're getting virtually no I/O performance at all, and it doesn't matter if it's Postgresql or bonnie++ doing the I/O. iostat would report that we'd be doing only 2MB/sec, and we'd be seeing 90%+ iowait percentages in atop or top. So our solution is to fix out I/O subsystem. We're replacing the expensive TLA SAN storage device with a low end Fibre Channel raid (which we were going to do anyways, as even at it's best, the TLA SAN wasn't impressive), upgrading from an old 2.4 linux kernel to a newer 2.6 kernel, changing the I/O Scheduler to deadline, and upgrading the server hardware. Some combination of the above solves the problem.
Hopefully no one has been lying awake worrying about this problem :-), but I wanted to close it out, and to leave a permanent record in the archives for the next person who has this problem.
Brian