Reloading confs

Logan Shaw lshaw at emitinc.com
Fri Aug 25 00:48:01 IST 2006


On Thu, 24 Aug 2006, Craig White wrote:
> On Thu, 2006-08-24 at 16:46 +0100, Colin Jack wrote:
>> I presume I need to restart MailScanner every time a modify the .conf
>> files?
>> At the moment I do this by restarting sendmail ... is there an easier
>> way? :)

> You shouldn't ever start sendmail (or restart sendmail) - you should
> only restart MailScanner which in turns stops/starts sendmail for you.

I've never really understood this.  As far as I know, the
dependency graph for proper operation looks like this:

 	overall-mail-flow -> mailscanner
 	overall-mail-flow -> sendmail

But as far as I know, the graph doesn't contain either of
these two relationships:

 	mailscanner -> sendmail
 	sendmail -> mailscanner

Put another way, sendmail can happily run with MailScanner
stopped.  It will just queue up messages in the incoming queue,
and will possibly chew on delivering any messages remaining
in the outgoing queue.  Things will never move from one queue
to another, but that isn't sendmail's job.

Likewise, it also seems like MailScanner can happily run with
sendmail stopped.  Any messages that are in the incoming
queue will get processed and moved to the outgoing queue.
No new messages will be placed in the incoming queue and
messages in the outgoing queue will get delivered anywhere,
but that's not MailScanner's job.

As a result, I don't understand why the two services are tied
together in one startup script.  In fact, it seems definitely
preferable to be able to restart mailscanner and leave sendmail
running.  In fact, this is exactly what I do.  When I make a
configuration change, I stop MailScanner ("bwahahaha...") but
leave sendmail running.  I can then make sure that if one of my
users has his MUA connected to sendmail and is in the middle of
sending a 5 MB attachment, that won't be disturbed.  The users
won't ever notice a broken connection or refused connection on
port 587.  Remote servers won't ever notice a refused connection
on port 25 and as a result try and hit my secondary MX.  It just
makes more sense to me to leave sendmail up and let it queue
messages if there is no reason to take that service down.
It lessens the impact of making mailscanner changes.

So is there something I'm missing?  Is there a reason why it
is the way it is?

   - Logan


More information about the MailScanner mailing list