zentol opened a new pull request, #21973: URL: https://github.com/apache/flink/pull/21973
The Sender/Receiver invokables assume that the receiver is run with parallelism greater or equal to the sender invokable. If this is not satisfied than there is no input gate to read from, causing read operations to fail. The test was using an uneven number of slots. This caused one invokable to run with p=5, the other with p=6. Depending on who gets this 6th slot the test would fail or pass. I don't think there's a well-defined order in which this should happen, so I attribute this to wrong assumptions in the test. For clarity, with FLINK-30895 the adaptive scheduler assigns slots in order of maximum parallelism, and if 2 vertices have the same parallelism then by the hashcode of the slot sharing group. Previously it would assign them in whatever iteration order vertices were passed to the scheduler. -- 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]
