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

Posted to issues@flink.apache.org by "Yun Gao (Jira)" <ji...@apache.org> on 2022/04/13 06:28:06 UTC

[jira] [Updated] (FLINK-19895) Unify Life Cycle Management of ResultPartitionType Pipelined Family

     [ https://issues.apache.org/jira/browse/FLINK-19895?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Yun Gao updated FLINK-19895:
----------------------------
    Fix Version/s: 1.16.0

> Unify Life Cycle Management of ResultPartitionType Pipelined Family
> -------------------------------------------------------------------
>
>                 Key: FLINK-19895
>                 URL: https://issues.apache.org/jira/browse/FLINK-19895
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Runtime / Task
>            Reporter: Yuan Mei
>            Priority: Major
>             Fix For: 1.15.0, 1.16.0
>
>
> This ticket is to unify lifecycle management of `ResultPartitionType.PIPELINED(_BOUNDED)` and `ResultPartitionType.PIPELINED_APPOXIMATE`, so that we can get rid of the hacky attribute `reconenctable` introduced in FLINK-19693
>  
> In short:
> *The current behavior of PIPELINED(_BOUNDED) is* ->
>  Release partition as soon as consumer exits
>  Release partition as soon as producer fails/canceled
> *Current behavior of PIPELINED_APPOXIMATE* ->
>  Release partition as soon as producer fails/canceled
>  Release partition when the job exists
> *Unified Pipelined Family to*
>  Release partition when producer exits.
>  
> One more question:
> *whether we can unify Blocking + Pieliened Family to*
>  Producer release partition when producer fails/canceled
>  Release partition when the job exists



--
This message was sent by Atlassian Jira
(v8.20.1#820001)