kubernetes: ignored pod-eviction-timeout settings

What happened: I modified the pod-eviction-timeout settings of kube-controller-manager on the master node (in order to to decrease the amount of time before k8s re-creates a pod in case of node failure). The default value is 5 minutes, I configured 30 seconds. Using the sudo docker ps --no-trunc | grep "kube-controller-manager" command I checked that the modification was successfully applied:

kubeadmin@nodetest21:~$ sudo docker ps --no-trunc | grep "kube-controller-manager"
387261c61ee9cebce50de2540e90b89e2bc710b4126a0c066ef41f0a1fb7cf38   sha256:0482f640093306a4de7073fde478cf3ca877b6fcc2c4957624dddb2d304daef5                         "kube-controller-manager --address=127.0.0.1 --authentication-kubeconfig=/etc/kubernetes/controller-manager.conf --authorization-kubeconfig=/etc/kubernetes/controller-manager.conf --client-ca-file=/etc/kubernetes/pki/ca.crt --cluster-signing-cert-file=/etc/kubernetes/pki/ca.crt --cluster-signing-key-file=/etc/kubernetes/pki/ca.key --controllers=*,bootstrapsigner,tokencleaner --kubeconfig=/etc/kubernetes/controller-manager.conf --leader-elect=true --requestheader-client-ca-file=/etc/kubernetes/pki/front-proxy-ca.crt --root-ca-file=/etc/kubernetes/pki/ca.crt --service-account-private-key-file=/etc/kubernetes/pki/sa.key --use-service-account-credentials=true --pod-eviction-timeout=30s" 

I applied a basic deployment with two replicas:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: busybox
  namespace: default
spec:
  replicas: 2
  selector:
    matchLabels:
      app: busybox
  template:
    metadata:
      labels:
        app: busybox
    spec:
      containers:
      - image: busybox
        command:
        - sleep
        - "3600"
        imagePullPolicy: IfNotPresent
        name: busybox
      restartPolicy: Always

The first pod created on the first worker node, the second pod created on the second worker node:

NAME         STATUS   ROLES    AGE   VERSION
nodetest21   Ready    master   34m   v1.13.3
nodetest22   Ready    <none>   31m   v1.13.3
nodetest23   Ready    <none>   30m   v1.13.3

NAMESPACE     NAME                                 READY   STATUS    RESTARTS   AGE   IP          NODE         NOMINATED NODE   READINESS GATES
default       busybox-74b487c57b-5s6g7             1/1     Running   0          13s   10.44.0.2   nodetest22   <none>           <none>
default       busybox-74b487c57b-6zdvv             1/1     Running   0          13s   10.36.0.1   nodetest23   <none>           <none>
kube-system   coredns-86c58d9df4-gmcjd             1/1     Running   0          34m   10.32.0.2   nodetest21   <none>           <none>
kube-system   coredns-86c58d9df4-wpffr             1/1     Running   0          34m   10.32.0.3   nodetest21   <none>           <none>
kube-system   etcd-nodetest21                      1/1     Running   0          33m   10.0.1.4    nodetest21   <none>           <none>
kube-system   kube-apiserver-nodetest21            1/1     Running   0          33m   10.0.1.4    nodetest21   <none>           <none>
kube-system   kube-controller-manager-nodetest21   1/1     Running   0          20m   10.0.1.4    nodetest21   <none>           <none>
kube-system   kube-proxy-6mcn8                     1/1     Running   1          31m   10.0.1.5    nodetest22   <none>           <none>
kube-system   kube-proxy-dhdqj                     1/1     Running   0          30m   10.0.1.6    nodetest23   <none>           <none>
kube-system   kube-proxy-vqjg8                     1/1     Running   0          34m   10.0.1.4    nodetest21   <none>           <none>
kube-system   kube-scheduler-nodetest21            1/1     Running   1          33m   10.0.1.4    nodetest21   <none>           <none>
kube-system   weave-net-9qls7                      2/2     Running   3          31m   10.0.1.5    nodetest22   <none>           <none>
kube-system   weave-net-h2cb6                      2/2     Running   0          33m   10.0.1.4    nodetest21   <none>           <none>
kube-system   weave-net-vkb62                      2/2     Running   0          30m   10.0.1.6    nodetest23   <none>           <none>

To test the correct pod eviction I shutdown the first worker node. After ~1 min the status of the first worker node changed to “NotReady”, then I had to wait +5 minutes (which is the default pod eviction timeout) for pod on the turned off node to be re-created on the other node.

What you expected to happen: After the node status reports “NotReady”, the pod should be re-created on the other node after 30 seconds instead if the default 5 minutes!

How to reproduce it (as minimally and precisely as possible): Create three nodes. Init Kubernetes on the first node (sudo kubeadm init), apply network plugin (kubectl apply -f "https://cloud.weave.works/k8s/net?k8s-version=$(kubectl version | base64 | tr -d '\n')"), then join the other two nodes (like: kubeadm join 10.0.1.4:6443 --token xdx9y1.z7jc0j7c8g8lpjog --discovery-token-ca-cert-hash sha256:04ae8388f607755c14eed702a23fd47802d5512e092b08add57040a2ae0736ac). Add pod-eviction-timeout parameter to Kube Controller Manager on the master node: sudo vi /etc/kubernetes/manifests/kube-controller-manager.yaml:

apiVersion: v1
kind: Pod
metadata:
  annotations:
    scheduler.alpha.kubernetes.io/critical-pod: ""
  creationTimestamp: null
  labels:
    component: kube-controller-manager
    tier: control-plane
  name: kube-controller-manager
  namespace: kube-system
spec:
  containers:
  - command:
    - kube-controller-manager
    - --address=127.0.0.1
    - --authentication-kubeconfig=/etc/kubernetes/controller-manager.conf
    - --authorization-kubeconfig=/etc/kubernetes/controller-manager.conf
    - --client-ca-file=/etc/kubernetes/pki/ca.crt
    - --cluster-signing-cert-file=/etc/kubernetes/pki/ca.crt
    - --cluster-signing-key-file=/etc/kubernetes/pki/ca.key
    - --controllers=*,bootstrapsigner,tokencleaner
    - --kubeconfig=/etc/kubernetes/controller-manager.conf
    - --leader-elect=true
    - --requestheader-client-ca-file=/etc/kubernetes/pki/front-proxy-ca.crt
    - --root-ca-file=/etc/kubernetes/pki/ca.crt
    - --service-account-private-key-file=/etc/kubernetes/pki/sa.key
    - --use-service-account-credentials=true
    - --pod-eviction-timeout=30s

(the yaml is truncated, only the related first part is showed here).

Check that the settings is applied: sudo docker ps --no-trunc | grep "kube-controller-manager"

Apply a deployment with two replicas, check that one pod is created on first worker node, the second is created on the second worker node. Shut down one of the nodes, and check the elapsed time between the event, when the node reports “NotReady” and the pod re-created.

Anything else we need to know?: I experience the same issue in multi-master environment also.

Environment:

• Kubernetes version (use kubectl version): v1.13.3 Client Version: version.Info{Major:“1”, Minor:“13”, GitVersion:“v1.13.3”, GitCommit:“721bfa751924da8d1680787490c54b9179b1fed0”, GitTreeState:“clean”, BuildDate:“2019-02-01T20:08:12Z”, GoVersion:“go1.11.5”, Compiler:“gc”, Platform:“linux/amd64”} Server Version: version.Info{Major:“1”, Minor:“13”, GitVersion:“v1.13.3”, GitCommit:“721bfa751924da8d1680787490c54b9179b1fed0”, GitTreeState:“clean”, BuildDate:“2019-02-01T20:00:57Z”, GoVersion:“go1.11.5”, Compiler:“gc”, Platform:“linux/amd64”}
• Cloud provider or hardware configuration: Azure VM
• OS (e.g: cat /etc/os-release): NAME=“Ubuntu” VERSION=“16.04.5 LTS (Xenial Xerus)”
• Kernel (e.g. uname -a): Linux nodetest21 4.15.0-1037-azure #39~16.04.1-Ubuntu SMP Tue Jan 15 17:20:47 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
• Install tools:
• Others: Docker v18.06.1-ce