Karmada — Multi-Cloud & Multi-Cluster Kubernetes Orchestration

What it is

Karmada (Kubernetes Armada) is a CNCF-incubated project that provides a control plane for managing workloads across multiple Kubernetes clusters. It extends the Kubernetes API so you can deploy and manage applications across clouds using familiar kubectl commands.

Karmada targets platform teams running Kubernetes in multiple clouds or regions who need a unified control plane for workload distribution, failover, and resource scheduling.

How it saves time or tokens

Without Karmada, managing workloads across multiple clusters requires per-cluster kubectl contexts, separate deployments, and manual failover scripts. Karmada provides a single API surface with PropagationPolicy resources that declaratively distribute workloads based on cluster labels, weights, and availability.

How to use

1. Install the kubectl-karmada plugin and initialize:

kubectl krew install karmada
kubectl karmada init

1. Join member clusters:

kubectl karmada join member1 --cluster-kubeconfig=member1.kubeconfig

1. Create a PropagationPolicy to distribute workloads:

apiVersion: policy.karmada.io/v1alpha1
kind: PropagationPolicy
metadata:
  name: distribute-web
spec:
  resourceSelectors:
    - apiVersion: apps/v1
      kind: Deployment
      name: web
  placement:
    clusterAffinity:
      clusterNames:
        - member1
        - member2
    replicaScheduling:
      replicaSchedulingType: Divided

Example

# Deploy to multiple clusters with one command
kubectl apply -f deployment.yaml --kubeconfig ~/.kube/karmada-apiserver.config
kubectl apply -f propagation-policy.yaml --kubeconfig ~/.kube/karmada-apiserver.config

# Check workload distribution
kubectl get rb -o wide --kubeconfig ~/.kube/karmada-apiserver.config

Related on TokRepo

• DevOps Tools -- Kubernetes and cloud infrastructure tools
• Self-Hosted Tools -- Self-hosted orchestration platforms

Common pitfalls

• Karmada requires a dedicated control plane cluster. This cluster runs the Karmada API server and does not run application workloads.
• Network connectivity between the control plane and member clusters is required. Firewalls must allow the Karmada control plane to reach member cluster API servers.
• PropagationPolicy changes trigger rebalancing. Avoid frequent policy updates that could cause unnecessary workload migrations.