What is the difference between a headless service and other types of services?

A headless service does not have a cluster IP address and does not perform load balancing. Instead, DNS directly returns the IP addresses of all pods matching the service selector. This allows the client to choose which pod to connect to or to use its own load balancing algorithm.

Typical use cases:

• StatefulSets: For applications requiring unique network identification and stable data storage.
• Discovery: Clients can directly discover pods and manage connections.
• Custom load balancing: Applications can implement their own load balancing algorithms.

Comparison with ClusterIP and NodePort services:

Service Type	Cluster IP	Load Balancing	DNS
ClusterIP	Yes	Kubernetes	Returns service IP
NodePort	Yes	Kubernetes	Returns service IP
Headless	No	No (client)	Returns pod IP addresses

Definition of a headless service in YAML:

apiVersion: v1
kind: Service
metadata:
  name: my-headless-service
spec:
  selector:
    app: my-app # Selector for pods
  clusterIP: None # Key difference - no cluster IP
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8080

Instead of accessing my-headless-service via a single IP, the client will receive a list of pod IP addresses:

$ kubectl get pods -l app=my-app -o custom-columns=NAME:.metadata.name,IP:.status.podIP
NAME            IP
my-app-abcde    10.42.0.10
my-app-fghij    10.42.0.11
my-app-klmno    10.42.0.12

The client can then choose any of these IP addresses to establish a connection.