Arias Hung wrote:
On Thu, 08 Jun 2006, Daryl C. W. O'Shea delivered in simple text monotype:
As for the copy_config timeouts... what kind of system load are you
seeing. 10, 50, 500, or higher? The current 20 seconds alarm is
twice the original alarm timeout, but if you've got a high enough load
it could still be a problem. You could increase this value to
something practically unusable, like 300, but I'd be really surprised
(and would like to about) if the timeout isn't being caused by insane
load or excessive swapping.
Actually, with the load at just around 5-6 I'm notcing spam starting to
seep through. When my snapshot utility kicks in twice
a day to take snapshots of my filesystem that seems to unleash a the
largest torrents of unchecked spam.
fs snapshots would make swap effectively useless. You'll be waiting all
day for disk I/O.
So... how much memory do you have in this machine, how much is free,
and how much (hopefully none or little) swap is being used. If swap
is being used, how much of the spamd processes are being swapped out
(check will the system is idle after it's been busy for a bit).
1 gig. Here's a sample of my free/spam:
# free -m
total used free shared buffers cached
Mem: 1010 982 28 0 94 149
-/+ buffers/cache: 738 272
Swap: 1953 516 1436
If you look at 'top' I'm sure you'll see that a good portion of the
spamd processes have been swapped out whenever you see this happen.
This is with the alarm timeout value increased to 40 in spamd as you
suggested.
Yeah, that's not really going to help. It's going to take a LONG time
if you're swap thrashing.
Does this ring any bells?
Yup. I'm almost certain that this is definitely caused by the spamd
processes being swapped in and out.
Thanks for you continued help on this.
No problem. Bill is in the mail! :)
Daryl
