jmckenzie-dev commented on PR #2554: URL: https://github.com/apache/cassandra/pull/2554#issuecomment-1673619426
> <cost talking> Ok. So part of my reaction is the name `cost`; if we rename that to `split_size` I'm a lot more ok with it. But I'd think maybe we revise the calculation to something like: `num_splits = (test_count / (split_size)) * $AGENT_COUNT_SCALAR`, where `AGENT_COUNT_SCALAR` is a user-provided fudge-factor downstream to indicate that, for a given test, they have N agents they can call up. That'd let people parameterize dialing up and down parallelization based on their unique env (i.e. if you have double the agents that can run a unit test at `medium_executor` vs. what you have available for `large_executor`, having some way to provide that to the scripts / reference system would be nice. OR - we could just document that and let the users figure it out. But the num_splits = num_tests / hardcoded_value feels like it gives rise to a hardcoded value that is going to be env-specific. So maybe... hm. We provide the `default_tests_per_split` (or whatever) for each job and document that folks may want to adjust that, and the higher they drive that number the more they can expect the suite to parallelize if agents are available? -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
