Decommission a node in AWS EKS

This guide describes a recommended workflow for decommissioning a Portworx node in your cluster.

note

The following steps don't apply if you're using an auto-scaling group (ASG) to manage your Portworx nodes. For details about how you can change the size of your auto-scaling group, see the Scaling the Size of Your Auto Scaling Group page of the AWS documentation.

Migrate application pods using Portworx volumes that are running on this node

If you plan to remove Portworx from a node, applications running on that node using Portworx need to be migrated. If Portworx is not running, existing application containers will end up with read-only volumes and new ones will fail to start.

Perform the following steps to migrate select pods.

1. Cordon the node using the following command:

   kubectl cordon <node>

2. Reschedule application pods using Portworx volumes on different nodes:

   kubectl delete pod <pod-name> -n <application-namespace>

   Since application pods are expected to be managed by a controller like Deployment or StatefulSet, Kubernetes will spin up a new replacement pod on another node.

Decommission Portworx

To decommission Portworx, perform the following steps.

Remove Portworx installation from the node

Apply the px/enabled=remove label and it will remove the existing Portworx systemd service. It will also apply the px/enabled=false label to stop Portworx from running in future.

For example, below command will remove existing Portworx installation from minion2 and also ensure that Portworx pod doesn’t run there in future.

kubectl label nodes < node > px/enabled=remove --overwrite

note

Decommission from Kubernetes: If the plan is to decommission this node altogether from the Kubernetes cluster, no further steps are needed.

Ensure application pods using Portworx don’t run on this node

If you need to continue using the node without Portworx, you will need to ensure your application pods using Portworx volumes don’t get scheduled here.

You can ensure this by adding the schedulerName: stork field to your application specs (Deployment, Statefulset, etc). Stork is a scheduler extension that will schedule pods using Portworx PVCs only on nodes that have Portworx running. Refer to the Using scheduler convergence article for more information.

Another way to achieve this is to use inter-pod affinity

• You need to define a pod affinity rule in your applications that ensure that application pods get scheduled only on nodes where the Portworx pod is running.

• Consider the following nginx example:

  apiVersion: apps/v1
  kind: Deployment
   metadata:
    name: nginx-deployment
  spec:
    selector:
      matchLabels:
        app: nginx
    replicas: 1
    template:
      metadata:
        labels:
          app: nginx
      spec:
        affinity:
          # Inter-pod affinity rule restricting nginx pods to run only on nodes where Portworx pods are running (Portworx pods have a label
          # name=portworx which is used in the rule)
          podAffinity:
            requiredDuringSchedulingIgnoredDuringExecution:
            - labelSelector:
                matchExpressions:
                - key: name
                  operator: In
                  values:
                  - "portworx"
              topologyKey: kubernetes.io/hostname
              namespaces:
              - "kube-system"
        hostNetwork: true
        containers:
        - name: nginx
          image: nginx
          ports:
          - containerPort: 80
          volumeMounts:
          - name: nginx-persistent-storage
            mountPath: /usr/share/nginx/html
        volumes:
        - name: nginx-persistent-storage
          persistentVolumeClaim:
            claimName: px-nginx-pvc

  It can also be configured on cluster level by adding spec.stork.args.webhook-controller Set to true in StorageCluster to make Stork the default scheduler for workloads using Portworx volumes:

  apiVersion: core.libopenstorage.org/v1
  kind: StorageCluster
  metadata:
    name: portworx
    namespace: portworx
  spec:
    stork:
      enabled: true
      args:
        webhook-controller: true

Remove Portworx from the cluster

Follow this guide section "Removing offline Nodes" or "Removing a functional node from a cluster" to decommission the Portworx node from the cluster.

Uncordon the node

You can now uncordon the node using: kubectl uncordon <node>

If you want to permanently decommission the node, you can skip the following step.

(Optional) Rejoin node to the cluster

If you want Portworx to start again on this node and join as a new node, follow the node rejoin steps.