lostluck commented on a change in pull request #12934:
URL: https://github.com/apache/beam/pull/12934#discussion_r495217147
##########
File path: sdks/python/apache_beam/runners/worker/sdk_worker.py
##########
@@ -543,15 +602,19 @@ def process_bundle_split(self,
instruction_id # type: str
):
# type: (...) -> beam_fn_api_pb2.InstructionResponse
- processor = self.bundle_processor_cache.lookup(request.instruction_id)
- if processor:
- return beam_fn_api_pb2.InstructionResponse(
- instruction_id=instruction_id,
- process_bundle_split=processor.try_split(request))
- else:
+ try:
+ processor = self.bundle_processor_cache.lookup(request.instruction_id)
+ except RuntimeError:
return beam_fn_api_pb2.InstructionResponse(
- instruction_id=instruction_id,
- error='Instruction not running: %s' % instruction_id)
+ instruction_id=instruction_id, error=traceback.format_exc())
+ # Return an empty response if we aren't running. This can happen
+ # if the ProcessBundleRequest has not started or already finished.
+ process_bundle_split = (
+ processor.try_split(request)
+ if processor else beam_fn_api_pb2.ProcessBundleSplitResponse())
Review comment:
So to be clear, the semantics are required to be:
Receive PB request for INST. -> Empty Splits & and Progress
INST starts having results -> Return results.
INST finishes -> EmptySplits and progress until "expired".
And the "I don't know" is an error for every other circumstance?
In the case where the bundle finishes, but requires Finalization, it INST
should remain valid and return the it's current set of responses for splits and
progress as usual, rather than the empties.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
