[deanq]

What

Endpoint.run_sync and Job.output returned None when a job ended FAILED, TIMED_OUT or CANCELLED, and the job's error was dropped. A failed job looked the same as a handler that returned None:

endpoint.run_sync({"input": {...}})   # handler raised ValueError
# before: None
# after:  JobError: job <id> FAILED: {"error_type": "<class 'ValueError'>", "error_message": "...", ...}

Changes:

• New runpod.error.JobError(RunPodError) with job_id, status and error.
• Endpoint.run_sync, Job.output and the asyncio Job.output raise it for any final state other than COMPLETED. The asyncio docstring already said output() raises when the job fails.
• COMPLETED jobs return their output as before, including None.
• CANCELLED is now handled the same as the other final states in run_sync, which previously only checked COMPLETED, FAILED and TIMED_OUT.

Behavior change

Code that checked result is None to detect a failure now gets a JobError instead. Raising was chosen over returning the error so a failure can't be silently treated as a result. It matches runpod.apps, which raises RemoteExecutionError for failed jobs.

Test plan

• New tests/test_endpoint/test_failed_jobs.py:
  • run_sync raises with status and error for FAILED, TIMED_OUT and CANCELLED
  • run_sync raises when a polled job fails
  • a COMPLETED job with no output still returns None
  • Job.output raises on every call after a failure
  • a failure without an error field still names the status
  • asyncio Job.output raises on failure
• The raise tests failed before the change
• make quality-check (2781 passed)
• Live check against a deployed queue endpoint whose handler raises: on unfixed code run_sync returned None; with this change run_sync and run().output() raise JobError with status FAILED and the handler's error message. A successful call still returns its output.

[celine-runpod]

bugbot run

[Copilot]

🟡 Changes recommended

Both synchronous and asynchronous runners return None after streaming completes because the cached completed status bypasses fetching the available aggregate output from /status.

2 open issues

• Fetch missing output before returning completed job results · New
• Fetch missing aggregate output for completed streamed jobs · New

What changed in this PR

This PR makes endpoint job failures distinguishable from successful jobs that return None, exposing failure details through JobError.

Changes:

• Adds JobError with job ID, status, and error details.
• Raises it for failed, timed-out, and cancelled jobs in synchronous and asynchronous output retrieval.
• Adds regression tests for failure handling and successful None output.

File	Description
tests/​test_endpoint/​test_failed_jobs.py	Tests terminal failures, repeated error reporting, and successful None output.
runpod/​error.py	Defines the new JobError exception.
runpod/​endpoint/​runner.py	Caches failure details and raises errors from synchronous job results.
runpod/​endpoint/​helpers.py	Adds shared terminal-state error handling.
runpod/​endpoint/​asyncio/​asyncio_runner.py	Applies failure handling to asynchronous job output.

_{Review effort: Balanced}