Hey there! Let's break down how to safely restart those core kube-system pods in your K3s cluster. Since these pods are managed by Kubernetes controllers (DaemonSets or Deployments), we don't manually "stop" them—instead, we trigger a restart that the controller will handle to automatically recreate the pods.

Option 1: Restart Individual Pods #

If you need to restart a single specific pod (e.g., a misbehaving calico-node-xxxx), you can delete it directly. The associated controller will immediately spin up a replacement:

# Replace "calico-node-xxxx" with your actual pod name
kubectl delete pod calico-node-xxxx -n kube-system

You can get the full list of pod names with:

kubectl get pods -n kube-system

Option 2: Batch Restart All Pods of a Component #

For restarting all instances of a component (e.g., every calico-node across your cluster), use label-based deletion or controller rollouts. This is more efficient than deleting pods one by one.

Using Label-Based Deletion #

Each component has unique labels you can target:

• calico-node: k8s-app=calico-node
• calico-kube-controllers: k8s-app=calico-kube-controllers
• metrics-server: k8s-app=metrics-server
• local-path-provisioner: app=local-path-provisioner
• coredns: k8s-app=kube-dns

Run this command to delete all pods matching a label (the controller will recreate them automatically):

# Example: Restart all calico-node pods
kubectl delete pods -n kube-system -l k8s-app=calico-node

Using Rollout Restart (Recommended for Production) #

This method performs a rolling restart—it replaces pods one at a time, minimizing downtime. It's the safest approach for core cluster components:

• For DaemonSets (like calico-node):

  kubectl rollout restart daemonset calico-node -n kube-system
  

• For Deployments (all other listed components):

  kubectl rollout restart deployment calico-kube-controllers -n kube-system
  kubectl rollout restart deployment metrics-server -n kube-system
  kubectl rollout restart deployment local-path-provisioner -n kube-system
  kubectl rollout restart deployment coredns -n kube-system
  

Option 3: Automated Script for Full Restart #

If you want to restart all these components in one go, use this bash script. It uses rollout restarts (the safest method) and waits for pods to become ready:

#!/bin/bash

echo "Initiating rolling restart of core kube-system components..."

# Restart DaemonSets
kubectl rollout restart daemonset calico-node -n kube-system

# Restart Deployments
kubectl rollout restart deployment calico-kube-controllers -n kube-system
kubectl rollout restart deployment metrics-server -n kube-system
kubectl rollout restart deployment local-path-provisioner -n kube-system
kubectl rollout restart deployment coredns -n kube-system

echo "Waiting for all pods to reach ready state..."
# Wait for calico-related pods
kubectl wait --for=condition=ready pod -n kube-system -l k8s-app in (calico-node,calico-kube-controllers,metrics-server,kube-dns) --timeout=5m
# Wait for local-path-provisioner
kubectl wait --for=condition=ready pod -n kube-system -l app=local-path-provisioner --timeout=5m

echo "All core kube-system pods have been restarted and are ready!"

Important Notes #

• Timing: Restart these components during low-traffic periods to minimize impact on your workloads.
• Network Impact: Restarting calico-node will briefly interrupt network connectivity for pods on the affected node—plan accordingly.
• Verify State: After restarting, check pod status with kubectl get pods -n kube-system to ensure all are in the Running state.

内容的提问来源于stack exchange，提问作者solveit