DaemonSet

Run a pod on every node in the cluster automatically.

Time: ~10 minutes Difficulty: Beginner

What You Will Learn

• DaemonSet: exactly one pod per node, always
• Tolerations for running on control-plane nodes
• Mounting host paths (node-level logs, metrics)
• Rolling update strategy for DaemonSets
• Real-world use cases: log collectors, monitoring agents, network plugins

Deploy

Navigate to the demo directory:

cd demos/daemonset

kubectl apply -f manifests/namespace.yaml
kubectl apply -f manifests/node-monitor.yaml
kubectl apply -f manifests/log-collector.yaml

Verify

# One pod per node for each DaemonSet
kubectl get pods -n daemonset-demo -o wide

# See which nodes have pods
kubectl get daemonsets -n daemonset-demo

On minikube (single node), you will see one pod per DaemonSet. On a multi-node cluster, each DaemonSet would have one pod per node.

Check the Node Monitor

The node-monitor pod reports CPU, memory, disk, and container count every 30 seconds:

kubectl logs -l app=node-monitor -n daemonset-demo --tail=15

Check the Log Collector

The log-collector tails the most recent container log file on the node:

kubectl logs -l app=log-collector -n daemonset-demo --tail=10

What is Happening

manifests/
  namespace.yaml        # daemonset-demo namespace
  node-monitor.yaml     # ConfigMap + DaemonSet: reports node stats every 30s
  log-collector.yaml    # DaemonSet: tails container logs from the host

Key concepts:

Feature	How It Works
One pod per node	The DaemonSet controller schedules exactly one pod on each node
Tolerations	node-monitor tolerates the control-plane taint, so it runs there too
hostPath	Mounts /var/log/containers from the node filesystem into the pod
Rolling updates	maxUnavailable: 1 updates one node at a time
New nodes	When a node joins the cluster, the DaemonSet automatically schedules a pod on it

Experiment

1. Add a second minikube node and watch the DaemonSet auto-schedule:

   minikube node add
   kubectl get pods -n daemonset-demo -o wide -w

2. Remove the toleration from node-monitor.yaml and re-apply. On clusters with tainted control-plane nodes, the pod will be evicted from those nodes.

3. Restrict the DaemonSet to specific nodes using nodeSelector:

   spec:
     template:
       spec:
         nodeSelector:
           disk: ssd

Cleanup

kubectl delete namespace daemonset-demo

If you added a second node:

minikube node delete m02

Further Reading

See docs/deep-dive.md for a detailed explanation of DaemonSet scheduling, update strategies, node affinity vs nodeSelector, tolerations for special nodes, and common production patterns.

Next Step

Move on to HPA to learn about automatic horizontal scaling based on load.