We are running Oracle 8.1.7.4 and 9.2.0.5 on Red Hat Linux AS 2.1 with the 2.4.9-e.41enterprise kernel. We are have a problem where the database crashes and does not leave any trace files. It looks like some did a kill -9 on the pmon process. This happens randomly at night with no pattern of time or day or how often. Our IBM support contract was unable to provide any suggestions on how to debug from the OS level and Oracle cannot debug because it leaves no trace files. Does anyone know of anything that can be done to determine why the process is dying?

We first suspected that the backups were causing the problem. We disabled the backups and the problem continues to happen.

Any ideas would help, because I have no idea which direction to look.

I know that might sound overly simplistic, but have you gone through and looked at the cron set up for all the users on the box to see if anything might be kicking off on a regularly scheduled basis that might be causing the interference that you are running into? Be sure to check accounts other than root, even cron jobs by non-privileged users can have unexpected results.

If this does not turn up anything you might want to consider setting up a simple cron job that runs every minute that checks for the presence of your oracle processes, and when it is not detected launch off a few simple diagnostic snapshots such as….

top
cat /proc/meminfo
ps –ef
df –h

direct all of these to a text file that you can look at when you make it into the office. This might give you a little better insight to what was happening in the timeframe that oracle decided to take a dirt nap on you.