My Experienc About Kubernetes CNI Plugin Migration

Sat, 02 May 2026 00:00:00 +0000

Tip

TL;DR: If you’re a developer looking for alternatives and migration plans for CNI plugin on your Kubernetes cluster, Cilium is a good choise which supports live-migration from almost any CNIs, NetworkPolicies and comprehensive observability ecosystems.

CNI plugin migration for a running Kubernetes cluster, especially in production environment, is never a trivial task. One simple strategy is Blue/Green deployment, which provisions a new cluster with target CNI and moving workloads from current cluster to the new one. This approach ensures the least downtime(still need extra setup, e.g., loadbalancer, to ensure availability) and ignores discrepancy between CNIs. However, what if in-place migration is necessary? This is not common but challenging senario, and mostly needed when operating a cluster with stateful or unmanaged workloads. For instance, a platform provides computing resource by Kubernetes but operators have no control to services deployed on it. If operators want to migrate CNI without letting users aware/involve, in-place migration will still be required.

Kubernetes Dynamic Client Requires Plural Resource Type

Fri, 01 May 2026 00:00:00 +0000

In GoLang, in addition to invoking kubectl, an alternative approach to access CRDs on a Kubernetes cluster via Dynamic client. It allows users to access CRDs programmatically without exec module and kubectl, which requires handling shells, exit code and STDIN/STDOU. However, it seems the module is not well document. An general error type users encounter often is resource type not found related error, which could be triggered by the code:

func GetNetworkPolicy(ctx context.Context, client dynamic.Interface, policy string) error {
	gvr := schema.GroupVersionResource{
        Group:    "cilium.io",
        Version:  "v2",
        Resource: "ciliumclusterwidenetworkpolicy",
    }

    policy, err := client.Resource(gvr).Get(ctx, policy, metav1.GetOptions{})
    if err != nil {
        return fmt.Errof("Error getting policy: %v", err)
    }
	// ... snipped
}

The issue here is the resource type. I didn’t find documents mention that directly, but value ofResource member in the GroupVersionResource structure should be plural. To make code above work correctly, ciliumclusterwidenetworkpolicy should be ciliumclusterwidenetworkpolicies. This is different from writing K8s objects in yaml format which singular is used. By contrary, if the problem is target objects not found, then users should get errors like ciliumclusterwidenetworkpolicy not found.

Kubernetes on Hallblazzar: Dev Journal

My Experienc About Kubernetes CNI Plugin Migration

Kubernetes Dynamic Client Requires Plural Resource Type