Discovering DaemonSets in Kubernetes

When using Deployments, you have to run a workload on each node of your cluster explicitly. A more experienced colleague has told you that you should use a DaemonSet instead. You want to know what a DaemonSet is and how it works.

With ReplicaSets, you can specify how many replicas of a given Pod should be running in your cluster. However, you cannot determine on which nodes the Pods are running. Kubernetes, as a container orchestrator, decides on which nodes the Pods should be scheduled based on the available resources on the nodes. However, in some cases, there must be only one copy of the Pod running on a given set of nodes in your cluster.

A DaemonSet is a unique Kubernetes workload that ensures that a given Pod runs on a subset of nodes in your Kubernetes cluster. Typically, DaemonSets are helpful for services such as logging agents, monitoring agents, or other services that need to run as a single copy on a subset of nodes in your cluster.

Workloads defined as DaemonSets are not scheduled by the default scheduler kube-scheduler. Instead, the DaemonSet controller creates a Pod on each node by default. A subset of nodes can be selected by specifying a node selector. The DaemonSet controller then creates a Pod on each node that matches the specified node selector.

The controller ensures that the Pod is running on the specified node. If the node is deleted, the controller creates a new Pod on the new node. Like the kube-scheduler, the DaemonSet controller also runs the reconciliation loop to ensure that the desired state is always met and reschedules the Pod if necessary.

See a simple DaemonSet manifest that declares the fluentd logging agent:

1234567891011121314151617
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: fluentd-daemon
spec:
  selector:
    matchLabels:
      app: fluentd-daemon
template:
   metadata:
     labels:
       app: fluentd-daemon
   spec:
      containers:
      - name: fluentd-daemon
        image: fluent/fluentd
...

There is no explicit node selector specified in this example. As a result, the DaemonSet controller schedules a Pod on all worker nodes.

By adding a node selector, you define a selector of labels assigned to the nodes. The DaemonSet controller then creates a Pod on each node that labels match the specified selector.

12345678910111213141516171819
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: fluentd-daemon
spec:
  selector:
    matchLabels:
      app: fluentd-daemon
  template:
    metadata:
      labels:
        app: fluentd-daemon
    spec:
      nodeSelector:
        run-fluentd: "true"
      containers:
      - name: fluentd-daemon
        image: fluent/fluentd
...

Nodes labeled via kubectl label nodes <node-name> run-fluentd=true are selected by the node selector, and the DaemonSet controller schedules a Pod on each of these nodes.

In this lesson, you learned about the differences between ReplicaSets and DaemonSets. You also learned that by adding a node selector to the DaemonSet manifest, you ensure that a given Pod runs on a subset of nodes in your cluster.