Promise.get(timeout, unit) throws a misleading TimeoutException when the workflow is canceled

[antmendoza]

Expected Behavior

Actual Behavior

cancelling a workflow that is blocked in Promise.get(timeout, unit) ends the wait immediately and reports it as a timeout

Steps to Reproduce the Problem

Attached you can find the reproduction

ExternalCancelRacingChildStartTest.java

Specifications

• Version: temporal sdk 1.38
• Platform:

[Quinn-With-Two-Ns]

Hmm yes this appears to be a long standing bug.

Root cause:

1. Promise.get(timeout, unit) calls the non-cancellable path in temporal-sdk/src/main/java/io/temporal/internal/sync/CompletablePromiseImpl.java:82.

2. Its timeout is implemented by WorkflowInternal.await(...), which creates a workflow timer.

3. That timer is attached to the current cancellation scope in temporal-sdk/src/main/java/io/temporal/internal/sync/SyncWorkflowContext.java:1001.

4. External workflow cancellation therefore cancels and completes the timer before its one-hour deadline.

5. Timed await only checks timer.isCompleted() and cannot distinguish “fired” from “canceled” (v1.38.0 source
   (https://github.com/temporalio/sdk-java/blob/v1.38.0/temporal-sdk/src/main/java/io/temporal/internal/sync/SyncWorkflowContext.java#L1328-L1357)).

6. The child-start promise remains incomplete, so CompletablePromiseImpl converts the early wake-up into TimeoutException (v1.38.0 source
   (https://github.com/temporalio/sdk-java/blob/v1.38.0/temporal-sdk/src/main/java/io/temporal/internal/sync/CompletablePromiseImpl.java#L81-L105)).