Swap used up, then oom-killer kills Apache 2 on Debian 4.0
So, I have a serious problem.
Quick question: Can segmentation fault in Apache2 or OCI8 module consume swap?
Long question:
I have Debian 4.0 on Dell PowerEdge 2950. It runs Apache2 with OCI8, and an application written in Java. A few users are testing a web app, and at seemingly random times, swap usage jumps to 8GB (full), then oom-killer kills apache instances. Now, we have exactly the same web app and Java app on a different machine, which is a normal PC but with Debian 5.0, and swap usage is next to 0, and RAM usage is much lower too. Also, on machine1 OCI8 version was 1.3.4, and on machine2 OCI8 is 1.3.5.
I read in OCI release notes that new version fixed a bug where it caused segmentation fault if ORACLE_HOME variable was not set. Of course, it was not set here, so I set it to correct value. The server now seems to behave more politely, but still swap usage is slowly increasing, although I haven't managed to "DOS" it with multiple requests (which was quite doable earlier).
I suppose that the best thing to do would be to install 5.0, but I am NOT the admin; he is on vacation, so it will have to wait. As I said, things look better now, but I'd like to know if it is possible that the segfaults have eaten up swap?
Also, Java app doesn't seem to be the issue, since it actually does something 2 times a day, and in that time it uses very little RAM, while Apache2 has 10+ instances which use 5-12%MEM (as top command reports).
Any ideas?
Thanx in advance..
|