[Mimedefang] The -y option

[Joseph Brennan]

Verified in production: running without -y definitely solves the problem in
the situation as described.

The mimedefang process is now taking between 0.1 and 0.3 %CPU. It grew to
well over 50% with -y.

No one else has noticed this? It's a huge difference.

Joseph Brennan
Columbia University Information Technology




--On June 22, 2016 at 08:57:38 -0400 Joseph Brennan <brennan at columbia.edu>
wrote:

> 
> As we investigated a problem with memory usage on our new (6 weeks) mail
> servers we noticed that the "mimedefang" process (the binary) grows larger
> over time (days), even larger than 4 GB. We took a close look at "what
> changed" and we noticed that Redhat (release 6.7) standard
> /etc/init.d/mimedefang slips in the -y option, which we never used. 
> 
> The man page mentions "This function leaked memory in versions of
> Sendmail prior to 8.14.4 so by default we do not call it". Hmm.
> 
> We have Redhat's rpm of Sendmail 8.14.4 and Mimedefang 2.78, so it seems
> all right. With our mods it starts up as:
> 
> /usr/bin/mimedefang -P /var/spool/MIMEDefang/mimedefang.pid -m
> /var/spool/MIMEDefang/mimedefang-multiplexor.sock -y -R -1 -U defang -a
> auth_author -a client_ptr -a client_resolve -q -p
> /var/spool/MIMEDefang/mimedefang.sock
> 
> It looks like it is still leaking, at least with these Redhat rpms. We did
> not compile so I can't verify what mods they may have done or that the
> milter library is the one from 8.14.4.
> 
> I'm going to run without -y on a test host and verify that we get the
> sendmail macros we want without it. I expect we do--we always did. We may
> need to run with real mail load for a few days to confirm that removing -y
> solves the problem.
> 
> Noted for comment and for advice to others.
> 
> 
> Joseph Brennan
> Columbia University Information Technology
>