[Enhancement]: KafkaRebalance Tools

[kornys]

Related problem

KafkaRebalance CRs drive Cruise Control rebalancing operations. They go through states (New → PendingProposal → ProposalReady → Rebalancing → Ready/NotReady) and provide optimization results. Essential for capacity planning and pre-upgrade checks.

Suggested solution

New Tools

list_kafka_rebalances

• Parameters: namespace (optional), clusterName (optional)
• Returns: List of rebalances: name, namespace, cluster, mode (full/add-brokers/remove-brokers/remove-disks), current state (from conditions), auto-approval status
• K8s CRD: kafkarebalances.kafka.strimzi.io

get_kafka_rebalance

• Parameters: namespace (optional), rebalanceName (required)
• Returns: Detailed rebalance: mode, goals, excluded topics, broker list, concurrency settings, replication throttle, optimization result (data movement summary, load before/after), session ID, progress reference, conditions with timestamps
• Note: Optimization result can be large — use ResponseSizeLimitFilter naturally

New Prompt

assess-upgrade-readiness

• Parameters: cluster_name, namespace (optional), target_version (optional)
• Workflow:
  1. Get Kafka cluster — check current versions (kafkaVersion, kafkaMetadataVersion)
  2. List node pools — verify all replicas ready, no pending changes
  3. Get Kafka metrics — check for under-replicated partitions (must be 0)
  4. Get operator status — verify operator is healthy and correct version
  5. List rebalances — check no active rebalancing in progress
  6. Get events — check for recent reconciliation failures
  7. Assess: all brokers healthy? Replication caught up? No pending operations? Operator ready?

New Files

• KafkaRebalanceService.java — @ApplicationScoped
• KafkaRebalanceTools.java — @Singleton, @Guarded
• KafkaRebalanceSummary.java / KafkaRebalanceDetail.java — DTOs
• AssessUpgradeReadinessPrompt.java
• KafkaRebalanceServiceTest.java, KafkaRebalanceToolsTest.java
• Update McpDiscoveryTest.java, PromptCompletions.java

RBAC

Add to 003-ClusterRole.yaml:

- apiGroups: ["kafka.strimzi.io"]
  resources: ["kafkarebalances"]
  verbs: ["get", "list", "watch"]

Key Notes

• KafkaRebalance state is encoded in conditions, not a dedicated status field. Extract state from condition types.
• Optimization result is a JSON structure within the CR status — extract key metrics (data movement GB, replica movements, leader movements).
• The strimzi.io/rebalance annotation (approve/stop/refresh) is write-only — we never set it, only report its current value.

Verification

1. mvn clean test — new tests pass
2. Create KafkaRebalance CRs in various states on dev cluster
3. Invoke list_kafka_rebalances, get_kafka_rebalance — verify state and optimization data
4. Test assess-upgrade-readiness prompt against a healthy and unhealthy cluster