I noticed that the cilium pods are generously provisioned on a DO k8s cluster. It seems a little excessive provision 30% of the CPU with no upper limit. Given it’s a Go project, and they’re usually quite performant, this reasonable?
kube-system cilium-crswr 300m (30%) 0 (0%) 0 (0%) 0 (0%) 10d
Hi there,
With cilium usage like this can be expected. Cilium provides the software defined network for our DOKS clusters. The reason we dont put an upper limit on cilium is because if the cilium pod goes down all workloads running on that node will lose network connectivity. Therefore capping the cilium pod’s resources at a limit would cause kubenetes to kill the cilium pod if it tried to take more than its limit causing an outage anyway.
It’s for this reason that cilium does not have a cap as its a crucial infrastructure component that is in our customers best interest to give it the resources it needs to maintain a stable cluster.
Regards,
John Kwiatkoski
Senior Developer Support Engineer