[erlang-questions] centos 6.5 OOM kills heart process making restart impossible

[Michael Santos]

On Mon, Oct 20, 2014 at 03:19:30PM +0400, Anton Lebedevich wrote:
> Hello.
> 
> When erlang process (erlang-17.3-2.el6 from Erlang Solutions packages)
> eats too much memory under CentOS 6.5 it is killed by OOM killer which
> is expected.
> But when I run it with -heart the heartbeat process gets killed first
> because it's a child of beam.smp which has the largest OOM score. It
> makes erlang application stop forever because both beam.smp and heart
> are dead.
> I don't remember similar problems under R14-16 and Debian.
> 
> Is there any way to make sure that beam.smp is killed first to allow
> heat process to restart it?

Disable the OOM killer. Add to /etc/sysctl.conf:

    vm.overcommit_memory=2
    vm.overcommit_ratio=90
    
The OOM killer can also be disabled per process. For example, if the
PID of heart is 12345:

    sudo echo -1000 > /proc/12345/oom_adj_score

    cat /proc/12345/oom_adj_score
    -1000

See proc(5) for details.