[Mimedefang] MXCommand: connect: Connection refused: Is multiplexor running?

[Nik Clayton]

Hi all,

We're seeing repeated instances of

    MXCommand: connect: Connection refused: Is multiplexor running?
    [...]
    mfconnect: Error communicating with multiplexor

on our content filtering farm.  Once MD starts logging these it never
seems to recover, and while it still processes some mail, has to be
restarted to get the logging back under control and process mail at normal
speed.

>From walking through the source code, we're getting as far as receiving
the SMTP client connection, and Sendmail's initialising the filter (the
start of srvrsmtp.c:smtp()).  The filter function, mfconnect() calls down
to MXCommand() (via MXCheckFreeSlaves()), which tries to connect(), and
fails.

The failure bubbles back up to Sendmail, which sets the tempfail flag, and
responds to the client's SMTP verbs with 4xx errors.

So far, so understood.

What I don't understand is why this is happening.  We start
mimedefang-multiplexor with '-I 50' to specify a ridiculously high socket
backlog.  If I run "md-mx-ctrl load" while this is happening the
multiplexor reports somewhere between 5 and 7 messages being processed per
second -- nowhere near the 50 backlog limit.

Rooting around Google has turned up a few similar report, with the
suggestion to increase the listen backlog.  Which we've done.

All this is with MD 2.48 on Solaris 8.

Any thoughts?  Does this sound familiar to anyone?

N