xintongsong commented on PR #21447: URL: https://github.com/apache/flink/pull/21447#issuecomment-1354188912
> Do you mean that we get the result of subtask level from job-vertex-cache in the first step, right? And don't trigger any thread sampling when get the result of subtask level, right? I meant, regardless of operator / subtask level, we always use job-vertex cache, and always trigger sampling of all subtasks if the target vertex is not cached. The only difference would be which samples we use to assemble the flame graph, all or of a specific subtask. > In the second step, we get the result of subtask level from job-vertex cache and execution-vertex cache, and choose the newer one, right? And trigger the subtask thread sampling when the execution-vertex cache is empty or stale, right? Yes. It's just there should not be a newer execution-vertex cache if there's already a valid job-vertex cache. So checking the job-vertex cache first practically means choosing the newer one. Alternatively, you can also update both job-vertex cache and execution-vertex cache when retrieving samples of the entire operator. In this way, you can only check the execution-vertex cache upon a subtask level query. -- 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: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
