Re: [cobalt-users] Sendmail dying

["Robert G. Fisher" <rfisher@xxxxxxxxxxxxxxx>]

A warning, I don't have a RaQ3, so I'm not familiar with either 
the version of sendmail, the linux kernel, or even the assortment 
of processes running on the machine.  However, I have had a little 
'fun' debugging sendmail crashing, and yes, I still have hair left ;-).

I would however recommend running 2.2.14 or higher, there were
some issues with the impelmentation of poll() in earlier 2.2.X
kernels that could result in blocking calls to syslog that will
result in all newly spawned sendmail processes halting as the
syslogd doesn't know how to answer their pleas for logging...

First off, What do you mean by 'sendmail crashes' -- If you login
via telnet when it is apparently crashed, and type:
	ps auxw | egrep [s]endmail | less

Does it show any active sendmail processess?  If so, how many and 
look at their status and the 'command line' space from the ps output.  

If for some reason, there are no sendmail processes running then 
I would suspect you are having ram issues.  Try running:
	nohup vmstat -n 30 > /tmp/vmstat.log &

That will redirect the output of vmstat into a file and let you know
how much memory is free along with how many processes are running or
swapped at any given moment -- the '30' will cause vmstat to output
a new line with updated information every 30 seconds...This should
help you determine if the memory is being completely consumed which
could starve the daemon and perhaps cause it to terminate abnormally.

Second off, what does your maillog show you?  Do you see any trends
at the periods it locked up?

Third, you might want to check your /etc/sendmail.cf file for a
few possible gotchas:

The following line will cause the parent sendmail process to report
an error to new connections after so many sendmail processes have
already been spawned.  
	O MaxDaemonChildren=#

The following line will cause sendmail to try and autorebuilt it's
alias database every 10 minutes or so, I think the default is True,
so I would suggest making sure you set it to false.
	O AutoRebuildAliases=True
		make this:
	O AutoRebuildAliases=False

This can be particuarly troublesome if you have a script that
periodically has to update this via a call to newaliases such
as the way the RaQ's management scripts tend to do.  What may
well happen is you can get into a race lock in which no one
wins until all sendmail processes are terminated -- and this
usually requires a SIGTERM or SIGKILL at this stage and restarted. 
This will show up on an strace btw.

Also check out the lines dealing with load average:

	O QueueLA=X
	O RefuseLA=Y

If your load is higher than the first, sendmail won't worry about
trying to send it but will dump it to the queue -- if it's higher
than Y, well then it will not answer connections.

Also keep an eye for the throttle for connections per second...
	O ConnectionRateThrottle=X


On Mon, May 22, 2000 at 03:05:02PM -0400, Grant Cummings wrote:
> Looked at top...but the issue remains with sendmail and not with qpop.  Pop
> services still function even when sendmail crashes.  Thanks though.
> 
> Any other suggestions?
> *********************
> Grant Cummings
> Support Manager
> Wholesaleisp.net
> *********************
> 
> 
> -----Original Message-----
> From: Adam Williams [mailto:adam@xxxxxxxxxxxxxx]
> Sent: Monday, May 22, 2000 2:31 PM
> To: 'cobalt-users@xxxxxxxxxxxxxxx'
> Subject: Re: [cobalt-users] Sendmail dying
> 
> 
> I had a similar problem.  THe problem was that qpopper was using 100% of
> the CPU at all times and overheating the processors and well as using all
> the resources.  I recompiled qpopper 3.0.1 by hand and replaced the binary
> and all has been fince since.  run top and see whats up on the system.
> 
> 			Adam
> 
> Respect my .Xauthoritah!
> _____________________________________________________________
> |Adam Williams                  adam@xxxxxxxxxx             |
> |Network Administrator          Vicksburg Chemical Company  |
> -------------------------------------------------------------
> 
> On Mon, 22 May 2000, Grant Cummings wrote:
> 
> > We have a RaQ3 with approximately 85 vhosts on it all of which have mail.
> > Sendmail has just been dying about once a day recently.  I called Cobalt
> > support and they believe it's under too heavy a load (why they believe
> this
> > is beyond me).
> > 
> > Are there any logs on the server that can point a finger as to why
> sendmail
> > dies?  When it dies it usually takes 15 minutes or so to restart too.  Is
> > that normal?  I've dug through the archives for this list and haven't
> found
> > a similar issue.  Maybe I'm just lucky!
> > 
> > Thanks,
> > *********************
> > Grant Cummings
> > Support Manager
> > Wholesaleisp.net
> > *********************
> > 
> > _______________________________________________
> > cobalt-users mailing list
> > cobalt-users@xxxxxxxxxxxxxxx
> > To Subscribe or Unsubscribe, please go to:
> > http://list.cobalt.com/mailman/listinfo/cobalt-users
> > 
> 
> 
> _______________________________________________
> cobalt-users mailing list
> cobalt-users@xxxxxxxxxxxxxxx
> To Subscribe or Unsubscribe, please go to:
> http://list.cobalt.com/mailman/listinfo/cobalt-users
> 
> _______________________________________________
> cobalt-users mailing list
> cobalt-users@xxxxxxxxxxxxxxx
> To Subscribe or Unsubscribe, please go to:
> http://list.cobalt.com/mailman/listinfo/cobalt-users

-- 
Robert G. Fisher		     NEOCOM Microspecialists Inc. 
System Administrator/Programmer      (540) 666-9533 x 116