So far, we've been using the Deployment object to deploy our Elasticsearch service. The Deployment Controller will manage the ReplicaSets and Pods under its control and ensure that the correct numbers are running and healthy.

However, a Deployment assumes that each instance is stateless and works independently from each other. More importantly, it assumes that instances are fungible—that one instance is interchangeable with any other. Pods managed by a Deployment have identical identities.

This is not the case for Elasticsearch, or other distributed databases, which must hold stateful information that distinguishes one Elasticsearch node from another. These Elasticsearch nodes need individual identities so that they can communicate with each other to ensure data is consistent across the cluster.

Kubernetes provides another API Object called StatefulSet. Like the Deployment object, StatefulSet manages the running and scaling of Pods, but it also guarantees the ordering and uniqueness of each Pod. Pods managed by a StatefulSet have individual identities.

StatefulSets are similar to Deployments in terms of definition, so we only need to make minimal changes to our manifests/elasticsearch/deployment.yaml. First, change the filename to stateful-set.yaml, and then change the kind property to StatefulSet:

kind: StatefulSet

Now, all the Pods within the StatefulSet can be identified with a name. The name is composed of the name of the StatefulSet, as well as the ordinal index of the Pod:

<statefulset-name>-<ordinal>