MailScanner memory spikes

[Edward L. Hannaford]

Just to add one more fact to the fire:

The spikes don't happen every 12 hours to the minute.  They actually happen
around 12-13 hours apart.  That's still consistent enough to be remarkable,
but not enough to help me! :-(

I was actually in the office this morning when a spike hit and I was able
to do a little bit of analysis.  Here are some fact snippets:

1) As before, memory utilization, both including and excluding cache,
spiked.  I managed to spit off some ps -fel commands to text files.  I
found that a MailScanner process had one of the biggest size reports
(whatever that size stat means).

2) The CPU utilization spiked also.  I can report that most of the
utilization was user, not system, CPU.

3) Here's the really weird part (to me, anyway).  To keep this server from
dragging to a halt I've configured MS for only one child.  If I ps ax I can
see the two MS processes: one parent, one child.  However, periodically, a
third MailScanner process will appear.  It normally goes away after a short
while, but it still shows up as a third process.  Its PID is not reported
anywhere in maillog, either (where normally a child process logs that it's
starting up).  This is consistent, by the way, with the behaviour of MS
when the max children was set to 5: normally I'd see 6 MS processes, but
every so often that would increase; moreover, during the spikes and
slowdowns, I'd see a total of 11 MS processes running!
   Even more odd, today when I ran ps -fel during the spike I discovered
that this extra, third MS process was a "grandchild!"  The MS parent
process hadn't spawned an extra child, but rather the MS child had spawned
its own child.  Moreover, it appeared to be this process which was
consuming inordinate amounts of memory and CPU.
   FWIW, these "grandchildren" appear regularly throughout the day and
don't spike anything.  I've got another message asking about these extra MS
processes.

-Ed