You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Xintong Song (Jira)" <ji...@apache.org> on 2021/08/10 10:31:00 UTC

[jira] [Commented] (FLINK-23707) Use consistent managed memory weights for StreamNode

    [ https://issues.apache.org/jira/browse/FLINK-23707?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17396599#comment-17396599 ] 

Xintong Song commented on FLINK-23707:
--------------------------------------

Thanks for bringing this up, [~twalthr].

I think it makes great sense and is important to keep the definition of "weight" consistent, at least within the same {{ManagedMemoryUseCase}}.

IIRC, the reason we didn't introduce such an explicit definition was that, table planner was the only user of the {{declareManagedMemoryUseCaseAtOperatorScope}}, so we sort of left the decision to the planner. That condition is no longer meet with both Table and DataStream sharing the {{ManagedMemoryUseCase#OPERATOR}}. I think it's a good idea to keep the definition of "weight" consistent, potentially together with the declaration of the operator scope use cases (i.e., in {{ManagedMemoryUseCase}}).


> Use consistent managed memory weights for StreamNode
> ----------------------------------------------------
>
>                 Key: FLINK-23707
>                 URL: https://issues.apache.org/jira/browse/FLINK-23707
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Table SQL / Planner
>            Reporter: Timo Walther
>            Assignee: Timo Walther
>            Priority: Major
>
> Managed memory that is declared on transformations via {{Transformation#declareManagedMemoryUseCaseAtOperatorScope(ManagedMemoryUseCase managedMemoryUseCase, int weight)}} should be declared using a weight.
> Usually, a weight should be some kind of factor, however, in the table planner it is used a kibi byte value. This causes issues on the DataStream API side that sets it to {{1}} in {{org.apache.flink.streaming.runtime.translators.BatchExecutionUtils#applyBatchExecutionSettings}}.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)