Rui Fan created FLINK-34504: ------------------------------- Summary: Avoid the parallelism adjutment when the upstream shuffle type doesn't has keyBy Key: FLINK-34504 URL: https://issues.apache.org/jira/browse/FLINK-34504 Project: Flink Issue Type: Improvement Components: Autoscaler Reporter: Rui Fan Assignee: Rui Fan Fix For: kubernetes-operator-1.8.0
JobVertexScaler#scale has a optimization: Try to adjust the parallelism such that it divides the number of key groups without a remainder => data is evenly spread across subtasks. It's only useful when the upstream shuffle type has keyBy. We should avoid this optimization when the upstream shuffle type doesn't has keyBy. -- This message was sent by Atlassian Jira (v8.20.10#820010)