On Mon, May 13, 2013 at 02:46:04PM -0600, Curtis wrote:
> Ok, we have confirmed that the postfix/smtp threads are not just
> hanging... after several minutes of logging nothing, each thread
> exits with a log entries that looks like this (real host names/IPs
> masked with ---):
>
> May 9 13:36:50 --- postfix/smtp[1114]: 3b3cyK07Bzz41vV6:
> conversation with ---.---.com[---.---.---.---] timed out while
> sending message body
> May 9 13:36:51 --- postfix/smtp[1114]: 3b3cyK07Bzz41vV6: enabling
> PIX workarounds: disable_esmtp delay_dotcrlf for
> ---.---.com[---.---.---.---]:25
> May 9 13:46:53 --- postfix/smtp[1114]: 3b3cyK07Bzz41vV6:
> to=<---@---.com>, relay=---.---.com[---.---.---.---]:25,
> delay=349260, delays=348054/0.01/604/602, dsn=4.4.2, status=deferred
> (conversation with ---.---.com[---.---.---.---] timed out while
> sending message body)
>
> ...with it being the same mail host that every active thread is
> trying to deliver to and eventually times out on.
If every smtp delivery agent is timing out sending mail to the same
destination, then part of the delay is due to all processes being busy
waiting for deliveries to that destination to complete. Have you set
a high destination concurrency limit for this transport? It is time to
share your main.cf (postconf -n) and master.cf settings.
If all the smtp(8) processes exit at essentially the same time, and
previously all were busy, you could see the behaviour you describe when
all mail goes to this same destination. For that you need to fix the
pix firewall, it is broken.
> So, if qmgr is still running, then my question remains the same...
> since the active queue is growing what are possible reasons why new
> smtp threads would not be spawning until every last active thread
> gives up on this non-responsive mail server?
Postfix has process limits, destination concurrency limits, ... If
enough gunk ties up all available resources the system will appear
to freeze.
If only the output is blocked, you'll indeed new mail entering the active
queue (qmgr from=... logging) while no output deliveries take place.
Do you see this or not? Time to share some log samples showing typical
activity before, during and after the "freeze"
--
Viktor.
