Kasia Kucharczyk created BEAM-5777:
--------------------------------------
Summary: Running ParDo in loop with DirectRunners raises
RuntimeException
Key: BEAM-5777
URL: https://issues.apache.org/jira/browse/BEAM-5777
Project: Beam
Issue Type: Bug
Components: testing
Reporter: Kasia Kucharczyk
Assignee: Jason Kuster
The Python [load test of ParDo operation for
SyntheticSources|https://github.com/apache/beam/blob/faff82860c66e4050f0cfa5e874ffe6035ed0c1c/sdks/python/apache_beam/testing/load_tests/par_do_test.py#L133]
that I created contains parametrized loop of ParDo with no operation inside
besides metrics (this issue). With setting the number of iterations to >~200
and running the test on DirectRunner I was encountering test failures. The test
outputs whole (really long) pipeline logs. Some test runs raised the following
exception:
{code:java}
Traceback (most recent call last):
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/testing/load_tests/par_do_test.py",
line 144, in testParDo
result = p.run()
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/testing/test_pipeline.py",
line 104, in run
result = super(TestPipeline, self).run(test_runner_api)
File "/Users/kasia/Repos/beam/sdks/python/apache_beam/pipeline.py", line 403,
in run
self.to_runner_api(), self.runner, self._options).run(False)
File "/Users/kasia/Repos/beam/sdks/python/apache_beam/pipeline.py", line 416,
in run
return self.runner.run_pipeline(self)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/direct/direct_runner.py",
line 139, in run_pipeline
return runner.run_pipeline(pipeline)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/portability/fn_api_runner.py",
line 229, in run_pipeline
return self.run_via_runner_api(pipeline.to_runner_api())
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/portability/fn_api_runner.py",
line 232, in run_via_runner_api
return self.run_stages(*self.create_stages(pipeline_proto))
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/portability/fn_api_runner.py",
line 1015, in run_stages
pcoll_buffers, safe_coders).process_bundle.metrics
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/portability/fn_api_runner.py",
line 1132, in run_stage
self._progress_frequency).process_bundle(data_input, data_output)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/portability/fn_api_runner.py",
line 1388, in process_bundle
result_future = self._controller.control_handler.push(process_bundle)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/portability/fn_api_runner.py",
line 1260, in push
response = self.worker.do_instruction(request)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/sdk_worker.py",
line 212, in do_instruction
request.instruction_id)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/sdk_worker.py",
line 231, in process_bundle
self.data_channel_factory)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/bundle_processor.py",
line 343, in __init__
self.ops = self.create_execution_tree(self.process_bundle_descriptor)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/bundle_processor.py",
line 385, in create_execution_tree
descriptor.transforms, key=topological_height, reverse=True)])
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/bundle_processor.py",
line 320, in wrapper
result = cache[args] = func(*args)
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/bundle_processor.py",
line 368, in get_operation
in descriptor.transforms[transform_id].outputs.items()
File
"/Users/kasia/Repos/beam/sdks/python/apache_beam/runners/worker/bundle_processor.py",
line 367, in <dictcomp>
for tag, pcoll_id
... (3 last lines repeated for long period)
RuntimeError: maximum recursion depth exceeded
{code}
>From my observation, I can say the problem appeared with various iteration
>number depending on computer resources. On my weaker computer started failing
>on ~150 iterations. The test succeeds on DataFlow with 1000 iterations (I
>didn't check higher number).
I can provide whole test output but it's ~1,3Mb.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
