2

I have a long-running process that is hitting a resource limit, such as the maximum number of open files.

I don't want to kill it.

Usually, you'd do:

(stop service)
ulimit -n <new limit>
(start service)

Is there a way to avoid having to stop and start the service and increase the limits?

3

I've figured it out.

On some kernels (e.g. 2.6.32+), at least on CentOS/RHEL, you can change the resource limits of a running process using /proc/<pid>/limits, e.g.:

$ grep "open files" /proc/23052/limits
Limit                     Soft Limit           Hard Limit           Units     
Max open files            1024                 4096                 files     

To change the maximum open files to a soft limit of 4096, hard limit of 8192:

echo -n "Max open files=4096:8192" > /proc/23052/limits

This gives:

$ grep "open files" /proc/23052/limits
Limit                     Soft Limit           Hard Limit           Units     
Max open files            4096                 8192                 files

Note the -n in echo -n - without that, you'll get an "invalid argument" error.

The above doesn't always work, so

Another option is prlimit command, introduced with util-linux 2.21 allows you to read and change the limits of running processes.

This is a followup to the writable /proc/<pid>/limits, which was not integrated in mainline kernel. This solution should work.

$ prlimit  --nofile --output RESOURCE,SOFT,HARD --pid 23052
RESOURCE SOFT HARD
NOFILE   1024 4096

Set the limits:

$ prlimit --nofile=4096:8192 --pid 23052

Confirm:

$ prlimit  --nofile --output RESOURCE,SOFT,HARD --pid 23052RESOURCE SOFT HARD
NOFILE   4096 8192
$ grep "open files" /proc/23052/limits
Limit                     Soft Limit           Hard Limit           Units     
Max open files            4096                 8192                 files     

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.