Re: [exim] Timeouts reading from spamd socket

On Thu, 2009-02-12 at 09:05 -0200, Eduardo M KALINOWSKI wrote:
> However, I still get the errors in my panic log (and an e-mail from a
> cron job) about that. I've found the log_selector option, but I couldn't
> find a way to disable only messages such as

I don't really think you can. The panic log is used to indicate really
serious conditions only, and is the ideal place to log things like SA or
AV timeouts or other problems.

Instead of deciding not to log them, or to only put them in the main log
(where they'll get swamped in all the other info), you should - ideally
- be addressing the problem of *why* SA took so long.

I would surmise that, if you look in the log SA generates
(probably /var/log/maillog but you may have it configured differently)
that you'll see something interesting about the message(es) which
trigger the error. Pay particular attention to the "child states"
messages.

Ideally you want to see:

Feb 12 11:00:08 bill spamd[28853]: prefork: child states: III

where I indicates "idle". What you really don't want to see too often
is:

Feb 12 11:28:10 bill spamd[6916]: prefork: child states:
BBBBBBBBBBBBBBBBBBB

where the number of Bs is the same as your "-m" option to spamd at
startup. That would indicate you're not running enough child processes.

If you're at the default setting, you can only process 5 simultaneous
messages so you may need to push that setting up accordingly.

Alternatively you may have a DNS check that's taking too long. But most
of this is for SA, not Exim, so I'll shut up now :)

Graeme