Pipeline lifecycle

Pipeline metadata and the Kafka Connect runtime are separate systems. ClueCDC stores both desired and last observed state; the UI never infers connector state from labels or a single successful HTTP request.

stateDiagram-v2
  [*] --> DRAFT: create and validate
  DRAFT --> RUNNING: deploy
  RUNNING --> PAUSED: pause
  PAUSED --> RUNNING: resume
  RUNNING --> RUNNING: restart connector/task
  PAUSED --> RUNNING: restart
  RUNNING --> DEGRADED: task failure observed
  DEGRADED --> RUNNING: runtime recovers
  DRAFT --> [*]: delete metadata
  RUNNING --> [*]: explicitly delete connector and metadata

Creation validates Debezium configuration but does not deploy. Deployment first provisions required topics, creates the connector, then commits runtime metadata; failure is explicit and compensating cleanup is attempted where safe. The worker periodically maps Kafka Connect connector/task state to ClueCDC domain state.

Those are separate API stages. The web wizard orchestrates them into one flow: capture preview, pipeline save, capture deployment, topic preparation, delivery preview, and first-delivery deployment. If the delivery stage fails, the already-created capture remains visible and the UI offers delivery retry.

Deleting a pipeline is explicit. It does not delete Kafka topics, messages, source tables, destination tables, replication publications, or slots. Kafka topic deletion is a separate API/UI operation with a permanent-data-loss warning. Deleting a destination is blocked while deliveries reference it.