[Bug]: FAILED invocation without ErrorObject is completed as SUCCEEDED by local emulator

[zhongkechen]

Expected Behavior

An invocation response with explicit Status=FAILED must result in a failed durable execution.

If an error object is required by the invocation contract, a FAILED response without one should be rejected or converted to a generic failure. It must never be converted into a successful execution.

Actual Behavior

In aws_durable_execution_sdk_python_testing.executor.Executor._validate_invocation_response_and_store(), the FAILED branch passes the nullable response error to _complete_workflow():

case InvocationStatus.FAILED:
    self._complete_workflow(
        execution_arn, result=None, error=response.error
    )

_complete_workflow() then chooses the terminal status based on whether the error is present:

if error is not None:
    self.fail_execution(execution_arn, error)
else:
    self.complete_execution(execution_arn, result)

Consequently, this invocation response:

{
  "Status": "FAILED"
}

is recorded as ExecutionSucceeded.

This was observed through SAM Local with the Java SDK when waitForCondition exhausted its attempts inside a synchronous child context. The operation history was:

StepFailed (WaitForCondition)
ContextFailed (RunInChildContext)
InvocationCompleted
ExecutionSucceeded

The producing Java SDK issue is aws/aws-durable-execution-sdk-java#634.

Steps to Reproduce

The emulator can be tested directly with a handler/invoker that returns:

{
    "Status": "FAILED"
}

Alternatively:

1. Run a Java durable handler through SAM Local.
2. Inside synchronous runInChildContext, execute waitForCondition with a built-in wait strategy and a condition that always continues polling.
3. Allow the operation to exhaust its maximum attempts.
4. Inspect the durable execution history.

The step and child context are recorded as failed, but the execution is recorded as succeeded.

SDK Version

aws-durable-execution-sdk-python-testing current main as of 2026-08-19. The affected code is used by the durable execution emulator image run by SAM Local.

Python Version

3.13

Is this a regression?

No known working version.

Last Working Version

N/A

Proposed Fix

Make the explicit invocation status authoritative:

case InvocationStatus.FAILED:
    error = response.error or create_generic_failure_error()
    self.fail_execution(execution_arn, error)

Alternatively, reject a FAILED response whose error is missing. Do not route the FAILED branch through logic that infers success from error is None.

Add a regression test asserting that:

• InvocationStatus.FAILED with an error produces a failed execution;
• InvocationStatus.FAILED without an error also cannot produce a successful execution;
• a ContextFailed checkpoint does not itself force root failure when the handler legitimately catches it and returns SUCCEEDED.