Re: [HTCondor-users] Limiting memory used on the worker node with c-groups

[Thomas Hartmann <thomas.hartmann@xxxxxxx>]

tbh, I am not 100% sure - I interpret it, that the OOM handling is done 
by Condor.
My guess would be, that something like
  oom_kill_disable 0
  under_oom 1
should mean, that the kernel would send processes to sleep but not kill them

On 24/04/2020 11.10, Jean-Michel Barbet wrote:
> On 4/24/20 10:50 AM, Thomas Hartmann wrote:
>
> > As far as I understand [1], with something like
> > ÂÂ > cat 
> > /sys/fs/cgroup/memory/system.slice/condor.service/SLOT/memory.oom_control
> > ÂÂ oom_kill_disable 1
> > ÂÂ under_oom 0
> > should indicate, that the kernel itself is not killing or stopping 
> > processes (but might depend also on the parent oom settigns maybe??)
>
> Hi Thomas,
>
> Thank you very much, the reference is very interesting.
> This is also what I have :
>
> oom_kill_disable 1
> under_oom 0
>
> Is this expected that some processes in the job tree go into "D"
> (uninterruptable sleep actually) whith this setting ?
> I would expect not...
>
> JM

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature