The role of the headless service is to allow clients in the Kubernetes cluster to connect to the Cassandra cluster through a standard Kubernetes service instead of keeping track of the network identities of the nodes or putting a dedicated load balancer in front of all the nodes. Kubernetes provides all that out of the box through its services.

Here is the configuration file:

apiVersion: v1 
kind: Service 
metadata: 
  labels: 
    app: cassandra 
  name: cassandra 
spec: 
  clusterIP: None 
  ports: 
    - port: 9042 
  selector: 
    app: Cassandra 

The app: Cassandra label will group all the pods to participate in the service. Kubernetes will create endpoint records and the DNS will return a record for discovery. The clusterIP is None, which means the service is headless and Kubernetes will not do any load balancing or proxying. This is important because Cassandra nodes do their own communication directly.

The 9042 port is used by Cassandra to serve CQL requests. Those can be queries, inserts/updates (it's always an upsert with Cassandra), or deletes.