39

I have a long running process that is eventually going to hit the max open file limit. I know how to change that after it fails, but is there a way to change that for the running process, from the command line?

Improve this question

5 Answers 5

Reset to default
40

As documented here, the 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.

If you don't have prlimit(1) yet, you can find the code to a minimalistic version in the prlimit(2) manpage.

Improve this answer
4
  • 2
    +1 for the minimalistic version in the manpage !
    – Totor
    Dec 14, 2017 at 10:39
  • Warning: "memcached" will segfault if you change "nolimit" on it's process after it encountered a limit. Not always a good idea to use the writeable limits
    – John
    Feb 9, 2019 at 1:31
  • OpenJDK 11 java is also segfaulted (but I only tried once).
    – Hontvári Levente
    Mar 24, 2020 at 11:41
  • Kernel doesn't provide interface for a process to be informed about changed limits. Some programs query limits during the process startup and may crash or misbehave if the limits are changed during the runtime. Increasing the limits at runtime should be okay for all correctly written programs, though.
    – Mikko Rantalainen
    Aug 23, 2021 at 8:34
33

On newer kernels (2.6.32+) on CentOS/RHEL you can change this at runtime with /proc/<pid>/limits:

cd /proc/7671/
[root@host 7671]# cat limits  | grep nice
Max nice priority         0                    0                    
[root@host 7671]# echo -n "Max nice priority=5:6" > limits
[root@host 7671]# cat limits  | grep nice
Max nice priority         5                    6
Improve this answer
6
  • That rules! Had no idea you could write to the limits file.
    – clee
    May 15, 2012 at 22:11
  • 4
    doesn't work for me on ubuntu 12.04
    – Poma
    Jul 10, 2012 at 7:06
  • 7
    This does not work on my 3.2 kernel. I guess your distribution has a specific unofficial patch for this because I see no trace of this patch in the kernel's fs/proc/base.c.
    – Totor
    Apr 19, 2013 at 13:43
  • Could be... I know it's present in RHEL / CentOS kernels. It doesn't seem to be present in Debian kernels.
    – Sig-IO
    Nov 12, 2015 at 10:38
  • 3
    'setprlimit' is the future-proof and recommended way, see the next answer
    – Sig-IO
    Nov 12, 2015 at 10:51
4

On newer version of util-linux-ng you can use prlimit command, for more infomation read this link https://superuser.com/questions/404239/setting-ulimit-on-a-running-process

Improve this answer
2

You can try ulimit man ulimit with the -n option however the mag page does not most OS's do not allow this to be set.

You can set a system wide file descriptions limit using sysctl -w fs.file-max=N and make the changes persist post boot up in /etc/sysctl.conf

However I would also suggest looking at the process to see if it really needs to have so many files open at a given time, and if you can in fact close some files down and be more efficient in the process.

Improve this answer
2
  • The open file thing is a bug, but I haven't been able to figure it out. While I'd love to do so, production periodically crashes mid-day.
    – kāgii
    Nov 12, 2010 at 15:57
  • 2
    ulimit does not apply settings to running processes.
    – try-catch-finally
    Jun 19, 2017 at 8:13
0

The process could change its own soft limits if programmed to do so (or if you manage to hack it), but it can't raise its hard limits unless it has the CAP_SYS_RESOURCE capability. You can inspect the limits at runtime in /proc/$pid/limits .

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .