TisonKun commented on issue #10451: [FLINK-15086][client] Explictly close JobCluster in attach mode URL: https://github.com/apache/flink/pull/10451#issuecomment-562442058 Verified locally | | without patch | with patch | |-|--------------|-------------| | execute + detach | cluster correctly closed | cluster correctly closed | | execute + attach | cluster correctly closed | cluster correctly closed | | executeAsync + detach | cluster correctly closed | cluster correctly closed | | executeAsync + attach | **cluster doesn't closed** | cluster correctly closed | actually here we have some statements. 1. executeAsync + detach: cluster will close itself on job finish, so the job client might cannot query the cluster before JobClient closed. but since it somehow means attach execution because you still communicate the job with job client but config `detach`, it is an undefined behavior. In this case, the recommended practice would be: fire `env.executeAsync` and forget it. 2. executeAsync + attach: 1) resource leak if user doesn't call `JobClient#close` in main method. OK we document it and highlight it. Now that the ownership of the cluster move to user program, he should take care of it. 2) cluster shutdown before job finished. It happens if user doesn't wait for job result but close job client before job finished. The job client is the handle you talk with the job; if you release it, the job is an orphan so close it would be OK. TBH for streaming workload which doesn't have a result anyway, it doesn't matter we just fire & forget. And close it when needed.
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
