Kubernetes Headless Service

Kubernetes Services are a fundamental abstraction that defines how pods are accessed within a cluster. While most people are familiar with ClusterIP, NodePort, and LoadBalancer services, Headless Services serve a very different and powerful purpose.

This article covers what a Kubernetes Headless Service is, why it exists, how it works, and when to use it

A Headless Service is a Kubernetes Service without a Cluster IP. Instead of a single virtual IP and load balancing, it provides direct pod IP access via DNS.

You create a headless service by setting:

spec:
  clusterIP: None

When this is done:

• Kubernetes does not allocate a virtual IP.
• Kube-proxy does not perform load balancing.
• DNS returns pod IPs directly, not a single service IP.

Traditional services abstract away individual pods and provide load balancing. However, some applications need to know exactly which pod they are talking to. There are some scenarios where more fine grained controlled is needed on which pod we want to connect for a specific need.

Common reasons can include:

• Stateful applications (databases, message brokers, caching systems)
• Leader–follower architectures
• Custom client-side load balancing
• Peer discovery systems

Headless services enable service discovery without traffic routing.

A DNS query in normal Service

my-service.default.svc.cluster.local

returns a service IP:

10.96.0.15

Traffic is load-balanced across pods.

DNS query:

my-headless-service.default.svc.cluster.local

Returns:

10.244.1.5
10.244.2.8
10.244.3.12

Each IP corresponds to a pod, not the service.

The client has the control to decide:

• Which pod to connect to
• How to load-balance or route traffic

To create a headless service, clusterIP needs to be set to None in the definition of a clusterIP server. Here is an example :

apiVersion: v1
kind: Service
metadata:
  name: my-headless-service
spec:
  clusterIP: None
  selector:
    app: my-app
  ports:
    - port: 80
      targetPort: 8080

Key points:

• clusterIP: None makes it headless
• Pods matching the selector are published via DNS.
• No virtual IP is created.

Headless services are most commonly used with StatefulSets.

Why do we need it?

StatefulSets give each pod:

• A stable hostname
• A stable network identity
• Persistent storage

Example DNS names:

pod-0.my-headless-service.default.svc.cluster.local
pod-1.my-headless-service.default.svc.cluster.local

This is ideal for:

• Databases like Cassandra, MongoDB, MySQL
• Kafka and Zookeeper
• etcd clusters

Depending on the configuration, Kubernetes creates:

• A records → One per pod IP
• SRV records → For named ports

Example SRV lookup:

my-port.tcp.my-headless-service.default.svc.cluster.local

This allows advanced clients to discover:

• Pod IP
• Port
• Protocol

Use a headless service when:

• You need direct pod-to-pod communication.
• Clients must discover all pods.
• You want custom or client-side load balancing.
• You're running stateful or clustered applications.

Do not use it when:

• You just want simple load balancing.
• You want a single stable service IP.
• The app does not need pod awareness.

• Full control over traffic routing.
• Essential for stateful workloads.
• Works seamlessly with StatefulSets.
• Enables advanced service discovery.

• No built-in load balancing.
• Client logic becomes more complex.
• Not suitable for simple stateless apps.

Author: https://medium.com/@narinderkaurmakkar1