https://rancher.com/ logo
Title
s

steep-window-46329

12/13/2022, 10:39 AM
Hello everyone. We are experiencing stability issues with downstream clusters where memory usage is high on worker nodes. This seems to be expected because there is no resource-reservation (
kube-reserved
,
system-reserved
) set per default on kubernetes and rancher. So there are evictions happening and in most cases just the OOM-Killer of Systemd running wild. We've set a custom value for
eviction-hard
and
system-reserved
and this seems to improve the stability of the worker-nodes and in case of low memory, the correct pods are getting killed instead of random system processes. Is there a general recommendation from Rancher for how much memory and CPU should be reserved on downstream-clusters based on the size of the worker-node? It would help a lot to get some general guidelines. Thanks! Reserve Compute Resources for System Daemons https://kubernetes.io/docs/tasks/administer-cluster/reserve-compute-resources/