Oom-killer - searching for a trigger

Linux Server
21

I don’t know their algorithm to decide what to kill. I think the tool is something like a last resort for avoiding the destruction of the system.

Setting user quotas probably could have a similar effect, but maybe in a better way. It would deny any further memory assignment to a particular user, ending up in destroying the user’s process, with potential data loss. I assume that a program consuming memory endlessly wouldn’t react gracefully to a memory denial.

2 Likes
22

Thank All of you for your answers and discussion.
I wrote before that I rebooted the server. Now is the second day and I hadn’t any problems. I didn’t expect that.
Like I wrote before, in log files I founded that every time oom-killer killed some other Process from different users. But I didn’t find the trigger. I am not going to search more.
I also bought 64GB of RAM and the second Processor.

3 Likes
23

This seems to be achievable in a way, according to the docs:

earlyoom v1.8
Usage: ./earlyoom [OPTION]...

  -m PERCENT[,KILL_PERCENT] set available memory minimum to PERCENT of total
                            (default 10 %).
                            earlyoom sends SIGTERM once below PERCENT, then
                            SIGKILL once below KILL_PERCENT (default PERCENT/2).
  -s PERCENT[,KILL_PERCENT] set free swap minimum to PERCENT of total (default
                            10 %).
                            Note: both memory and swap must be below minimum for
                            earlyoom to act.
  -M SIZE[,KILL_SIZE]       set available memory minimum to SIZE KiB
  -S SIZE[,KILL_SIZE]       set free swap minimum to SIZE KiB
  -n                        enable d-bus notifications
  -N /PATH/TO/SCRIPT        call script after oom kill
  -P /PATH/TO/SCRIPT        call script before oom kill
  -g                        kill all processes within a process group
  -d, --debug               enable debugging messages
  -v                        print version information and exit
  -r INTERVAL               memory report interval in seconds (default 1), set
                            to 0 to disable completely
  -p                        set niceness of earlyoom to -20 and oom_score_adj to
                            -100
  --ignore-root-user        do not kill processes owned by root
  --sort-by-rss             find process with the largest rss (default oom_score)
  --prefer REGEX            prefer to kill processes matching REGEX
  --avoid REGEX             avoid killing processes matching REGEX
  --ignore REGEX            ignore processes matching REGEX
  --dryrun                  dry run (do not kill any processes)
  --syslog                  use syslog instead of std streams
  -h, --help                this help text

So flag the process with a matching ignore REGEX for “dont kill me in any circumstances”.

5 Likes
24

That will help enormously. I have 64Gb ram… I can run several VM’s simultaneously, and it does not even blink. It is a 15 year old machine… it copes because it has ram to spare.

3 Likes