Wiring the SPIs and adding lifecycle control
The Worker Provisioner SPIs — WorkerStatusListener, WorkerContextProvider, CaseChannelProvider — had been defined with contract tests and no-op defaults. Nothing called them.
Wiring them in was mechanical in places but the call sites matter. onWorkerStarted goes in WorkerExecutionJobListener.jobToBeExecuted() — right when Quartz is about to fire the job, before execution begins. onWorkerCompleted in WorkflowExecutionCompletedHandler after output is applied to the context. onWorkerStalled in WorkerRetriesExhaustedEventHandler. buildContext() is called in WorkerScheduleEventHandler before scheduling — implementations can prepare worker context before the worker touches the job. openChannel fires on case start; listChannels plus closeChannel fire when the case reaches a terminal state.
For testing, we used @Alternative @Priority(1) static inner classes in @QuarkusTest — recording implementations that track invocation arguments, reset in @BeforeEach. No Mockito, no test profiles. In CDI 4.0, @Priority(1) activates the alternative globally across the entire test suite — the same behaviour that causes AmbiguousResolutionException when you do it accidentally becomes an asset when you do it deliberately with recording state that doesn’t change observable behaviour.
cancelCase, suspendCase, and resumeCase are now on CaseHub and CaseHubRuntime. State validation lives in CaseHubReactor — terminal states throw IllegalStateException, wrong-source transitions throw IllegalStateException, unknown caseIds throw IllegalArgumentException. Publishing to CASE_STATUS_CHANGED handles the rest; the existing handler takes care of persistence, scheduler cancellation, and channel cleanup.
Resume had one wrinkle. When a case transitions SUSPENDED → RUNNING, the engine doesn’t automatically know to re-evaluate which workers are eligible — it waits for a context change event. We added a CONTEXT_CHANGED publish in CaseStatusChangedHandler for that specific transition:
if (newState == CaseStatus.RUNNING) {
eventBus.publish(
EventBusAddresses.CONTEXT_CHANGED,
new CaseContextChangedEvent(caseInstance, caseInstance.getCaseContext().asJsonNode()));
}
One line. Without it, resume would be silent — the case moves to RUNNING but nothing fires until the next signal arrives. With it, any binding whose condition is already satisfied in the current context fires immediately.