[ https://issues.apache.org/jira/browse/FLINK-11299?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
shuai.xu reassigned FLINK-11299: -------------------------------- Assignee: shuai.xu > Decide the result partition type dynamically > --------------------------------------------- > > Key: FLINK-11299 > URL: https://issues.apache.org/jira/browse/FLINK-11299 > Project: Flink > Issue Type: Improvement > Components: JobManager > Reporter: shuai.xu > Assignee: shuai.xu > Priority: Major > > Now, the result partition type is decided when compiling JobGraph at the > client. This's to say, whether to output the result of a task to its > consumers by PIPELINED or BLOCKING is decided at client and can not be > changed any more. However, it is usually seen in batch jobs that a task is > configured PIPELINED but its consumers can not be started due to lack of > resource, this will lead to failover now. So we proposal to support deciding > the result partition type dynamically during runtime according to the > resource of cluster and so on to increase the success rate of batch job. -- This message was sent by Atlassian JIRA (v7.6.3#76005)