The systemReserved settings via kubeletConfig don't get enforced to the system daemons.

Customer experienced high load on a node on 30 May - "with the heavy load starting around 1900, we completely lost our node-exporter metrics at around 1925 and again at 20:40 for 5-10 minutes each."

Despite having KubeletConfig with systemReserved.cpu: 4000m, the cpu shares in /sys/fs/cgroup/cpu/system.slice/cpu.shares shows 1024.

Kubelet and CRI-O were unable to function.

4.12.57

100%

1. Set system-reserved via kubeletconfig
2. When the system is under high load, kubelet and crio don't function

The system daemons including kubelet and crio don't function under high load and the system-reserved doesn't get enforced

The system daemons including kubelet and crio should be working fine when system-reseved is set