Description
During shutdown, WorkerFactory asynchronously clears workerCommandWorker:
workerCommandWorker = null;
Meanwhile, awaitTermination checks the field and later dereferences it inside a lambda:
if (workerCommandWorker != null) {
long t = timeoutMillis;
ShutdownManager.runAndGetRemainingTimeoutMs(
t, () -> workerCommandWorker.awaitTermination(t, TimeUnit.MILLISECONDS));
}The field can become null between the check and lambda execution, producing:
java.lang.NullPointerException: Cannot invoke
"io.temporal.internal.worker.SuspendableWorker.awaitTermination(long, java.util.concurrent.TimeUnit)"
because "this.workerCommandWorker" is null
at io.temporal.worker.WorkerFactory.lambda$awaitTermination$12(WorkerFactory.java:519)
This was observed during normal AWS Lambda worker shutdown using SDK 1.37.0.
Expected behavior
awaitTermination should complete without an exception and consistently track the command worker’s termination.
Suggested resolution
A local capture prevents the immediate NPE but leaves timing-dependent lifecycle behavior.
The likely fix is to retain workerCommandWorker for the lifetime of WorkerFactory, as is done for ordinary workers, and remove the asynchronous null assignment. This also prevents isTerminated and awaitTermination from prematurely ignoring a command worker whose shutdown future completed before every underlying executor terminated.
If clearing the reference is intentional, it should use an explicit thread-safe lifecycle transition and stable termination primitive.
Environment
- Temporal Java SDK
1.37.0
DefaultLambdaWorkerRuntime- AWS Lambda
- Temporal Cloud
Introduced in sdk-java PR #2917.
Description
During shutdown,
WorkerFactoryasynchronously clearsworkerCommandWorker:Meanwhile,
awaitTerminationchecks the field and later dereferences it inside a lambda:The field can become
nullbetween the check and lambda execution, producing:This was observed during normal AWS Lambda worker shutdown using SDK
1.37.0.Expected behavior
awaitTerminationshould complete without an exception and consistently track the command worker’s termination.Suggested resolution
A local capture prevents the immediate NPE but leaves timing-dependent lifecycle behavior.
The likely fix is to retain
workerCommandWorkerfor the lifetime ofWorkerFactory, as is done for ordinary workers, and remove the asynchronousnullassignment. This also preventsisTerminatedandawaitTerminationfrom prematurely ignoring a command worker whose shutdown future completed before every underlying executor terminated.If clearing the reference is intentional, it should use an explicit thread-safe lifecycle transition and stable termination primitive.
Environment
1.37.0DefaultLambdaWorkerRuntimeIntroduced in sdk-java PR #2917.