Errors whenever mail is being processed

[David Crowe]

I am seeing a number of errors being generated to the monitor screen whenever we relay mail through sendmail. We have 2 HP 9000 boxes running sendmail with a couple of Interscan Virus servers running in between. The mail flows through the 1st Sendmail box and then through either one or the other Interscan boxes. Once the email starts to build up on the 2nd sendmail box (we are running a mail storm program that is sending approx 14,000 msg/hr)we start to see errors showing up on the monitor screen. The following are just a sample of the errors that are being generated "Feb 23 15:51:08 UX02 Sendmail[17135]: PAA17135: SYSERR(root): queueup: Cannot create temp file tfPAA17135, uid=0: Bad file number" and "Disconnect: Cannot open /dev/null: file table overflow". In the syslog.log I am seeing the error "vmunix: file: table is full". I have increased the number of open files and verified that we appear to have enough free drive space available but I still get this error. The 1st sendmail box has none of these errors and is actually handling more messages than the 2nd sendmail box. It is also a duplicate of the 2nd sendmail box.
I am wondering if something has gotten corrupted and if I should try to re-install Sendmail.

[Jitendra_1 · Accepted solution]

The "file table overflow " can be seen from sar -v . Check to see your nfile parameter. Increase it substantially to get rid of that file table error . I would increase nproc as well.

[Bill Hassell]

This is a common problem when a server is heavily loaded. There are table limits which must be significantly increased when the machine is stressed. For capacity planning, you can use sar but there can be peaks that are missed with the 1 second resolution of sar. HP-UX can run tens of thousands of programs at the same time and an average of 5 to 10 files will be opened at the same time.

Start with the worst case estimate of a mail storm, perhaps several hundred messages will be processed each second by a fast multi-processor system so nproc may need to be 5000 and nfile should be 18000 to 25000. Don't use the formula in SAM as it cannot predict loads caused by client/server processes (no logins).

[Steven Sim Kok Leong]

Hi,

Other than the nfiles kernel parameter, you may like to check your maxfiles kernel parameters as well. Apparently, your mailstorm program is generating a lot of concurrently open files.

To verify, you can perform an "lsof | grep mailstorm_process_name" to identify all the files opened by the process or processes.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com