Re: using zmailer for serious production

[smd@clock.org (Sean M. Doran)]

Tom Samplonius <tom@sdf.com> writes:

>   That cron line will cause bad things to happen.  It will launch a
> queueing running sendmail every minute.

I think that was the point.

>   As stated in the documentation, the correct way to keep the queue empty
> is by running sendmail as "sendmail -bd -q10m", which run the queue every
> 10 minutes.

The problem is that sendfail(8) tends to block on things,
and when blocked, -q<time> simply doesn't work.

So if you happen to have a series of messages that
overwhelm your bendmail(8)'s abilities to process them,
you end up with no mail delivery at all.

Zmailer is less prone to blocking on the head of the queue
than sendwhale(8) is, but it can still happen, as someone
else noted.  This is disappointing, but ameliorated
somewhat by reducing the most common cause for blocking,
namely long waits on the DNS resolver.   

Coughing up a time-limited DNS query engine that would let
one enqueue a message that didn't get a response of any
sort within a certain period of time, letting it be
processed when there is less pending work, would be a neat
feature.

Meanwhile, using a single large, reliable, topologically nearby
caching nameserver is a smart idea if one is running large
mailing lists and the like, no matter what MTA one is using.

>   Remember that in sendmail, messages only go into the queue, if initial
> delivery fails.

It can take a long time to get a NOERROR...

	Sean.