How to change the Container Runtime Interface (CRI) of an existing node pool

A container runtime is a software that, in addition to other tasks, ensures that containers are started and stopped accordingly. It is one the lowest layers of Kubernetes. Historically, Kubernetes used Docker as a container runtime. However, with ongoing development, it became clear that it was not entirely suitable. With the rapidly evolving world of container runtimes, other solutions emerged as better suited.
The Container Runtime Interface (CRI) was developed as a means to open up Kubernetes to other container runtimes. CRI is a plugin within a Kubernetes cluster that allows the kubelet to use more than one type of container runtime. It consists of a protocol buffer and a gRPC API, as well as libraries, with additional specifications and tools. At Scaleway, we favor Containerd, the industry-leading CRI solution. Containerd was initially developed by Docker and donated to the Cloud Native Computing Foundation (CNCF). It uses a subset of the original Docker engine and comes with most of Docker’s functionalities for running containers and managing storage and images - but lacks many developer-oriented features. It is, therefore, suitable for large-scale use as part of a container orchestrator, such as Kubernetes.

If you are still using one of the deprecated CRIs, your cluster will have the following limitations:

• Pools using deprecated CRIs cannot scale up.
• Control planes cannot be upgraded if the CRI used by a pool is deprecated.

To migrate your node pools to a supported CRI, you have to create a new resource pool supporting containerd, move your pods to the new pool and then delete your old pool.

How to move your pods to a new pool

1. Click Kubernetes in the Containers section of the Scaleway Console side menu. The Kubernetes Kapsule dashboard displays.
2. Click «See more Icon» next to the cluster you want to edit and click More info to find detailed information about the cluster.
3. Click the Pools tab to view the node pools associated with your cluster.
   Tip:

   Any node pool using docker or cri-o runtimes is deprecated.

4. Click + Add a new pool. The pool creation wizard displays.
5. Configure your new node pool:
   • Select a zone for your pool. The zone of your pool must match the zone your cluster is deployed in.
   • Select the node type to be used for your pool.
   • Configure pool options, such as autoscaling and autohealing.
   • Click Add pool to create the new pool and to add it to your cluster.
6. Use the kubectl get nodes command to retrieve a list of your nodes.
7. Mark the nodes of your old pool as unschedulable using the kubectl cordon command:

   kubectl cordon <old_node_1>
   kubectl cordon <old_node_2>
   ...Copy

   Important:

   If you have used local storage on your old nodes, data must be copied manually before moving pods and deleting nodes.

8. Emtpy the nodes using the kubectl drain command:

   kubectl drain <old_node_1>
   kubectl drain <old_node_2>
   ...Copy

   Your old nodes will be emptied and the pods are recreated in the new pool.
9. Delete your old pool by clicking on «See more Icon» > Delete next to the pool’s name once all your pods are up again.