remove potential mem overcommitment

[rdxmb]

When initializing the kubernetes-cluster with kubeadm, the resource-request and -limit for coredns differs. That can cause to mem overcommitment and oom-killed pods in the cluster.

In my case:

syslog:

[...]
[Mon Jul 13 04:07:15 2020] Memory cgroup out of memory: Kill process 48615 (mongod) score 1954 or sacrifice child
[...]

kubectl describe node drei
[...]
  kube-system                   coredns-66bff467f8-fcx9g               100m (5%)     0 (0%)      70Mi (0%)        170Mi (1%)       57d
[...]
  memory             16518324992 (98%)  16623182592 (99%)

My proposal is to set the limits to the same size as the requests.

[miekg]

this is the wrong repo for this request. And coredns org is probably also wrong.

[rdxmb]

Oh, I am sorry. What would be the right repo for that? @miekg

[chrisohaver]

What would be the right repo

It would be the repo for whatever tool you use to provision your cluster. More precisely, whatever tool defines the coredns Deployment.

[rdxmb]

More precisely, whatever tool defines the coredns Deployment.

Hm, not easy to say 🤔

It would be the repo for whatever tool you use to provision your cluster.

This is kubeadm init. I will see if I find the right repository for that ...

[chrisohaver]

It's kubeadm, every k8s deployment tool decides independently how to configure the CoreDNS deployment.

kubeadm lives in https://github.com/kubernetes/kubernetes.

[rdxmb]

@chrisohaver thanks 👍