Hello,
Today I had 1,500 sendmail processes in TCP CLOSE_WAIT state and
uninterruptible process state. All other network and non-network functions
appeared to function normally.
'ps auxwww' output:
USER PID %CPU %MEM VSZ RSS TTY S STARTED TIME COMMAND
root 391 0.0 0.0 2.78M 440K ?? U 13:28:09 0:00.01 sendmail:
startup with portal.gmu
All of the sendmail processes had the 'startup with HOST' message, many
different hosts, except for the parent sendmail that said 'rejecting
connections' due to the MaxDaemonChildren set. My understanding is the
processes were waiting for the application to close the connection according
to Comer and Stevens "Internetworking with TCP/IP." Has anyone else seen
this or know the cause? MaxDaemonChildren tells the sendmail daemon the
maximum number of children it is allowed to spawn.
I was able to kill the parent sendmail process and start another parent that
promptly spawned MaxDaemonChildren number of children in the hung state.
Killing the parent left many orphaned children. That is, after I killed the
parent sendmail and restarted a new daemon, I had 2 * MaxDaemonChildren
hung processes. I was forced to reboot the system to clear the situation. :(
Outbound email was still being processed but inbound was rejected, of
course. The problem began at the same time our router rebooted itself
- coincidence perhaps.
I am running sendmail 8.8.8 built and running on Digital UNIX 4.0B system
with patch kit #6, BL8. The system is a AlphaServer 4100 with 4 CPUs and
2GB of memory. This system processes anywhere from 100,000 to 1,000,000+
mail messages a day.
The same CLOSE_WAIT hung state began to happen to our SPSS license
manager, FlexLM, after upgrading to Digital UNIX 4.0B -- coincidence perhaps.
We did not have this problem under DU 3.2C, 3.2G.
--
Regards,
Richard Jackson
Computer Center Lead Engineer
Mgr, Central Systems & Dept. UNIX Consulting
University Computing & Information Systems (UCIS)
George Mason University, Fairfax, Virginia
Received on Wed Apr 08 1998 - 03:23:05 NZST