A DaemonSet ensures that all (or some) nodes in a cluster run a copy of a pod. DaemonSets are highly recommended for certain system-level applications, such as log collection and resource monitoring, because these applications run a pod on each node and do not require many pods. A good example of DaemonSet is the kube-proxy of Kubernetes.

DaemonSets are closely related to nodes. If a node crashes, the DaemonSet will not create another pod in other nodes.

Figure 1 DaemonSet

The following is an example of a DaemonSet:

apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: nginx-daemonset
  labels:
    app: nginx-daemonset
spec:
  selector:
    matchLabels:
      app: nginx-daemonset
  template:
    metadata:
      labels:
        app: nginx-daemonset
    spec:
      nodeSelector:                 #Node selection. A pod is created on a node only when the node meets daemon=need.
        daemon: need
      containers:
      - name: nginx-daemonset
        image: nginx:alpine
        resources:
          limits:
            cpu: 250m
            memory: 512Mi
          requests:
            cpu: 250m
            memory: 512Mi
      imagePullSecrets:
      - name: default-secret

The replicas parameter used in defining a Deployment or StatefulSet does not exist in the above configuration for a DaemonSet, because each node has only one replica. It is fixed.

The nodeSelector in the preceding pod template specifies that a pod is created only on the nodes that meet daemon=need, as shown in the following figure. If you want to create a pod on each node, delete the label.

Figure 2 DaemonSet creating a pod on nodes with a specified label

Create a DaemonSet.

$ kubectl create -f daemonset.yaml
daemonset.apps/nginx-daemonset created

Run the following command. The output shows that nginx-daemonset creates no pods on nodes.

$ kubectl get ds
NAME              DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
nginx-daemonset   0         0         0       0            0           daemon=need     16s
​
$ kubectl get pods
No resources found in default namespace.

This is because no nodes have the daemon=need label. Run the following command to query the labels of nodes:

$ kubectl get node --show-labels
NAME            STATUS   ROLES    AGE   VERSION                            LABELS
192.168.0.212   Ready    <none>   83m   v1.15.6-r1-20.3.0.2.B001-15.30.2   beta.kubernetes.io/arch=amd64 ...
192.168.0.94    Ready    <none>   83m   v1.15.6-r1-20.3.0.2.B001-15.30.2   beta.kubernetes.io/arch=amd64 ...
192.168.0.97    Ready    <none>   83m   v1.15.6-r1-20.3.0.2.B001-15.30.2   beta.kubernetes.io/arch=amd64 ...

Add the daemon=need label to node 192.168.0.212, and then query the pods of nginx-daemonset again. It is found that a pod has been created on node 192.168.0.212.

$ kubectl label node 192.168.0.212 daemon=need
node/192.168.0.212 labeled

$ kubectl get ds
NAME              DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
nginx-daemonset   1         1         0       1            0           daemon=need     116s

$ kubectl get pod -owide
NAME                    READY   STATUS    RESTARTS   AGE   IP           NODE            
nginx-daemonset-g9b7j   1/1     Running   0          18s   172.16.3.0   192.168.0.212

Add the daemon=need label to node 192.168.0.94. You can find that a pod is created on this node, as well.

$ kubectl label node 192.168.0.94 daemon=need
node/192.168.0.94 labeled

$ kubectl get ds
NAME              DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
nginx-daemonset   2         2         1       2            1           daemon=need     2m29s

$ kubectl get pod -owide
NAME                    READY   STATUS              RESTARTS   AGE   IP           NODE            
nginx-daemonset-6jjxz   0/1     ContainerCreating   0          8s    <none>       192.168.0.94    
nginx-daemonset-g9b7j   1/1     Running             0          42s   172.16.3.0   192.168.0.212   

Modify the daemon=need label of node 192.168.0.94. You can find the DaemonSet deletes its pod from the node.

$ kubectl label node 192.168.0.94 daemon=no --overwrite
node/192.168.0.94 labeled

$ kubectl get ds
NAME              DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
nginx-daemonset   1         1         1       1            1           daemon=need     4m5s

$ kubectl get pod -owide
NAME                    READY   STATUS    RESTARTS   AGE     IP           NODE            
nginx-daemonset-g9b7j   1/1     Running   0          2m23s   172.16.3.0   192.168.0.212