Cassandra K8s: Cassandra operator installation and lifecycle management

This domain tests the control loop: installing the operator and CRDs, declaring a CassandraDatacenter, and letting reconciliation own the StatefulSets, Services, and configuration.

Core rules

• The cass-operator watches CassandraDatacenter resources and reconciles generated objects toward the declared state.
• Install CRDs before the operator version that uses them; version skew is a common failure.
• The CR specifies serverType, serverVersion, size, racks, storageConfig, resources, and Cassandra configuration.
• Generated StatefulSets and Pods are not the source of truth; edits are overwritten on the next reconcile.
• Finalizers let the operator decommission nodes and clean up before a resource disappears.

Worked example

A CassandraDatacenter named cluster1 in namespace db declares size: 6 and three racks. The operator materializes rack-scoped StatefulSets with two replicas each, headless Services for peer discovery, and a client Service. If an administrator patches one generated StatefulSet to replicas: 3, reconciliation resets it to 2 because the CR still says six total. To add capacity, change size to 9; the operator creates one new node per rack, waits for readiness, and keeps seeds consistent.

Traps

• Installing the operator before matching CRDs.
• Deleting the CassandraDatacenter and assuming all PVCs vanish.
• Manually editing generated StatefulSets, ConfigMaps, or Services.
• Ignoring status conditions and proceeding while a rack is not ready.