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

Posted to issues@flink.apache.org by "Weijie Guo (Jira)" <ji...@apache.org> on 2022/11/29 12:48:00 UTC

[jira] [Updated] (FLINK-29769) Further limit the explosion range of failover in hybrid shuffle mode

     [ https://issues.apache.org/jira/browse/FLINK-29769?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Weijie Guo updated FLINK-29769:
-------------------------------
    Description: Under the current failover strategy, if a region changes to the failed state, all its downstream regions must be restarted. For ALL_ EDGE_BLOCKING type jobs, since they are scheduled stage by stage, no additional overhead. However, for the hybrid shuffle mode, the upstream and downstream can both run at the same time. If the upstream task fails, we hope that it will not affect the downstream tasks that do not consume it.  (was: Under the current failover strategy, if a region changes to the failed state, all its downstream regions must be restarted. For ALL_ EDGE_BLOCKING type jobs, since they are scheduled stage by state, no additional overhead. However, for the hybrid shuffle mode, the upstream and downstream can both run at the same time. If the upstream task fails, we hope that it will not affect the downstream regions that do not consume it.)

> Further limit the explosion range of failover in hybrid shuffle mode
> --------------------------------------------------------------------
>
>                 Key: FLINK-29769
>                 URL: https://issues.apache.org/jira/browse/FLINK-29769
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Runtime / Coordination
>    Affects Versions: 1.17.0
>            Reporter: Weijie Guo
>            Priority: Major
>
> Under the current failover strategy, if a region changes to the failed state, all its downstream regions must be restarted. For ALL_ EDGE_BLOCKING type jobs, since they are scheduled stage by stage, no additional overhead. However, for the hybrid shuffle mode, the upstream and downstream can both run at the same time. If the upstream task fails, we hope that it will not affect the downstream tasks that do not consume it.



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