kamilwu commented on a change in pull request #12435:
URL: https://github.com/apache/beam/pull/12435#discussion_r479204876
##########
File path: sdks/python/apache_beam/testing/load_tests/pardo_test.py
##########
@@ -125,7 +125,9 @@ def process(self, element, state=state_param):
state.add(1)
yield element
- if self.get_option_or_default('streaming', False):
+ if self.get_option_or_default(
+ 'streaming',
+ False) and self.pipeline.get_option('runner') == "PortableRunner":
Review comment:
I did some more tests in streaming on Flink and it turned out that
`SyntheticSource` is slower than `StatefulLoadGenerator`. I think it's because
`SyntheticSource` is not able to use the benefits of multiple workers (only one
task manager generates data).
That being said, I think we should keep two different sources for now. As
for the `if` statement above, in my opinion, a new switch (for example,
`--use_stateful_load_generator)` be a good idea.
----------------------------------------------------------------
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]
