Junrui Li created FLINK-32876: --------------------------------- Summary: ExecutionTimeBasedSlowTaskDetector treats unscheduled tasks as slow tasks and causes speculative execution to fail. Key: FLINK-32876 URL: https://issues.apache.org/jira/browse/FLINK-32876 Project: Flink Issue Type: Bug Components: Runtime / Coordination Affects Versions: 1.18.0 Reporter: Junrui Li Fix For: 1.18.0
When the ExecutionTimeBasedSlowTaskDetector determines slow tasks, it only requires that the ExecutionJobVertex is initialized and has not finished. However, when the SpeculativeScheduler schedules slow tasks, it requires that the corresponding ExecutionVertex has an ExecutionVertexVersion, which means that the ExecutionVertex has entered the scheduled state. This inconsistency can result in speculative execution failure. -- This message was sent by Atlassian Jira (v8.20.10#820010)