Hi,

I'm using a RHEL4U5 on many servers. (kernel 2.6.9-55)

I've 2 types of processes :
- 1 that is very important (Database)
- a multithreaded internal application that have a memory leak (in rare cases)

Sometimes, there were the condition for the memory leak, and it results in an OOM.

I know that the best solution is to debug the application (and developpers are working on it), but, for now, it isn't yet solved.

The oom killer choose to kill the database.

To prevent the database to be killed, i would like to adjust the oom score with /proc/PIT/oom_adj.

Does anyone know whether i can adjust oom on this kernel, or having an other solution?

Try it and see. I don't have a kernel that old, but if /proc/<pid>/oom_score exists, you'd have to think /proc/<pid>/oom_adj would work.
Have a look at ../Documentation/filesystems/proc.txt

oom_score doesn't exists. I'm afraid that i can't control the oom killer :-/
grep -i doesn't match 'oom' on proc.txt

Yeah, a quick check looks like it shipped in 2.6.20 ...

I saw it on a 2.6.12 kernel ;-) : http://www.ussg.indiana.edu/hypermai...04.0/0291.html

then 2.6.11 on man pages :
http://lwn.net/Articles/255421/

Thanks for helping me founding the answer