About MGRLU of oomkiller
I have this in my /etc/sysctl.d/99-oomkiller.conf
<span style="color:#323232;"># "Iron Reserve" that can not be consumed by rogue tasks
</span><span style="color:#323232;">vm.admin_reserve_kbytes = 1024000
</span><span style="color:#323232;">
</span><span style="color:#323232;"># 0 = extensive heuristic scan of joblist (system doesn't have the power for that, when oomkiller is needed). Can shoot the wrong task
</span><span style="color:#323232;"># 1 or more = kills the first task with condition "out of memory"
</span><span style="color:#323232;">vm.oom_kill_allocating_task = 1
</span>
Yet somehow i still got in a livelock. So i remembered nohang and found in it’s readme about mgrlu. I found some documentation to it but it only provides runtime examples (already enabled but set to 0 on Artix, /sys/kernel/mm/lru_gen/min_ttl_ms
should be 1000).
How should i set this permanently, sysctl doesn’t accept min_ttl_ms
. Via kernel, via local.d script (non-systemd, dinit)?
Why doesn’t my config above work?
Any recommendation to runtime-services? (earlyoom for my server i guess)
Add comment