You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@flink.apache.org by "Weijie Guo (Jira)" <ji...@apache.org> on 2022/08/25 03:01:00 UTC

[jira] [Created] (FLINK-29101) PipelinedRegionSchedulingStrategy benchmark shows performance degradation

Weijie Guo created FLINK-29101:
----------------------------------

             Summary: PipelinedRegionSchedulingStrategy benchmark shows performance degradation
                 Key: FLINK-29101
                 URL: https://issues.apache.org/jira/browse/FLINK-29101
             Project: Flink
          Issue Type: Bug
          Components: Runtime / Coordination
    Affects Versions: 1.16.0
            Reporter: Weijie Guo
             Fix For: 1.16.0


Throw TPC-DS and flink-benchmark testing, we found that PipelinedRegionSchedulingStrategy has performance degradation. By investigation, I can confirm that this was introduced by FLINK-28799 which introduce HYBRID type edge support for scheduling strategy.

The key to the problem is for blocking ALL_TO_ALL type edges should only enter the scheduling method when the last execution becomes finished, but the current implementation ignores this fact, resulting in the complexity of O(n ^ 2) in this case.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)