[
https://issues.apache.org/jira/browse/BEAM-3645?focusedWorklogId=279833&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-279833
]
ASF GitHub Bot logged work on BEAM-3645:
----------------------------------------
Author: ASF GitHub Bot
Created on: 19/Jul/19 16:53
Start Date: 19/Jul/19 16:53
Worklog Time Spent: 10m
Work Description: Hannah-Jiang commented on pull request #8979:
[BEAM-3645] add multiplexing for python FnApiRunner
URL: https://github.com/apache/beam/pull/8979#discussion_r305435276
##########
File path: sdks/python/apache_beam/runners/portability/fn_api_runner.py
##########
@@ -1480,42 +1664,54 @@ def process_bundle(self, inputs, expected_outputs):
# Register the bundle descriptor, if needed - noop if already registered.
registration_future = self._register_bundle_descriptor()
+ # Check that the bundle was successfully registered.
+ if registration_future and registration_future.get().error:
+ raise RuntimeError(registration_future.get().error)
split_manager = self._select_split_manager()
- if not split_manager:
- # If there is no split_manager, write all input data to the channel.
+ if not split_manager and isinstance(self._worker_handler,
+ EmbeddedWorkerHandler):
+ # If there is no split_manager, write all input data to the channel. For
+ # EmbeddedWorkerHandler, send data first, then send request.
for transform_id, elements in inputs.items():
self._send_input_to_worker(
process_bundle_id, transform_id, elements)
- # Check that the bundle was successfully registered.
- if registration_future and registration_future.get().error:
- raise RuntimeError(registration_future.get().error)
-
# Actually start the bundle.
process_bundle_req = beam_fn_api_pb2.InstructionRequest(
instruction_id=process_bundle_id,
process_bundle=beam_fn_api_pb2.ProcessBundleRequest(
process_bundle_descriptor_reference=self._bundle_descriptor.id))
- result_future = self._controller.control_handler.push(process_bundle_req)
+ result_future = self._worker_handler.control_handler.push(
+ process_bundle_req, self._worker_handler._worker_id)
+
+ if not split_manager and isinstance(self._worker_handler,
Review comment:
What I observed here is, with non-GRPC worker handler, if we send request
before sending data, bundle is treated as empty and processed immediately
before data is sent so it always return nothing, so we need to send data before
request. With GRPC worker handlers, request should be sent before data, because
we need to multiplex data according to the request. That's why I split it into
two parts, but I don't think this is an ideal way to handle this case.
----------------------------------------------------------------
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]
Issue Time Tracking
-------------------
Worklog Id: (was: 279833)
Time Spent: 26.5h (was: 26h 20m)
> Support multi-process execution on the FnApiRunner
> --------------------------------------------------
>
> Key: BEAM-3645
> URL: https://issues.apache.org/jira/browse/BEAM-3645
> Project: Beam
> Issue Type: Improvement
> Components: sdk-py-core
> Affects Versions: 2.2.0, 2.3.0
> Reporter: Charles Chen
> Assignee: Hannah Jiang
> Priority: Major
> Fix For: 2.15.0
>
> Time Spent: 26.5h
> Remaining Estimate: 0h
>
> https://issues.apache.org/jira/browse/BEAM-3644 gave us a 15x performance
> gain over the previous DirectRunner. We can do even better in multi-core
> environments by supporting multi-process execution in the FnApiRunner, to
> scale past Python GIL limitations.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
