`TaskOrchestrationContext.waitForExternalEvent` Timeout "Removed" After Attempting to Schedule Duplicate Orchestration

Summary

An orchestration which specifies a timeout for waiting on an external event will seemingly have that timeout "removed"/ignored after another attempt is made to schedule an orchestration instance with the same ID.

Steps to Reproduce

1. Copy the example below.
2. Execute the orchestrator-trigger function via HTTP.

@FunctionName("orchestrator-trigger")
public HttpResponseMessage triggerOrchestrator(
    @HttpTrigger(
        name = "request",
        methods = { HttpMethod.GET },
        authLevel = AuthorizationLevel.FUNCTION,
        route = "orchestrator/trigger"
    )
    HttpRequestMessage<String> request,
    @DurableClientInput(name = "durableContext")
    DurableClientContext durableContext,
    ExecutionContext context
) throws InterruptedException {
    DurableTaskClient client = durableContext.getClient();
    String instanceId = "the-only-instance";

    client.scheduleNewOrchestrationInstance("orchestrator", null, instanceId);
    client.raiseEvent(instanceId, "first", 1);

    // Attempting to schedule another orchestration instance with the same instance ID results
    // in the event timeout within the existing orchestrator instance from triggering.
    try {
        client.scheduleNewOrchestrationInstance("orchestrator", null, instanceId);
    } catch (RuntimeException ignored) { }

    Thread.sleep(10_000);
    client.raiseEvent(instanceId, "second", 2);

    return request.createResponseBuilder(HttpStatus.OK).build();
}

@FunctionName("orchestrator")
public void orchestrator(
    @DurableOrchestrationTrigger(name = "orchestration")
    TaskOrchestrationContext orchestration,
    ExecutionContext context
) {
    Task<Integer> firstTask =
        orchestration.waitForExternalEvent("first", Duration.ofSeconds(1), Integer.class);

    Task<Integer> secondTask =
        orchestration.waitForExternalEvent("second", Duration.ofSeconds(1), Integer.class);

    List<Integer> results = orchestration.allOf(firstTask, secondTask).await();
    int first = results.get(0);
    int second = results.get(1);

    System.out.printf("Triggered! First: %s, Second: %s\n", first, second);
}

Expected Result

The orchestration throws an exception due to the second event not arriving within one second.

Actual Result

After ten seconds, the orchestration prints Triggered! First: 1, Second: 2.

Additional Context

Deleting the entire try/catch block that contains the second scheduleNewOrchestrationInstance call results in the expected outcome - the orchestration throws.

This test case is contrived - my actual use case is:

• A process operates on the combination of a ZIP file and a CSV file.
• These two files are provided to the application separately, in any order, but at roughly the same time.
• Whichever file arrives first needs to start an orchestration instance that will wait for both files.
• We need to avoid a race condition that would result in two separate orchestration instances being created, each waiting for the opposite file. So we'll create a deterministic instance ID based on other data, have the CSV/ZIP receivers always attempt to schedule an orchestration instance with that ID, and ignore the "already exists" error when it occurs.
• The CSV/ZIP receivers then simply send their own single event to the orchestration instance.
• The orchestration instance needs to have a timeout in the situation where the second file never arrives.

Start with the Java reproduction in the issue, focusing on TaskOrchestrationContext.waitForExternalEvent and the duplicate scheduleNewOrchestrationInstance call. Run the orchestrator-trigger scenario and verify that the second external event times out after one second even when scheduling the same instance ID is attempted again.