[
https://issues.apache.org/jira/browse/SPARK-20079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16090854#comment-16090854
]
Apache Spark commented on SPARK-20079:
--------------------------------------
User 'vanzin' has created a pull request for this issue:
https://github.com/apache/spark/pull/18663
> Re registration of AM hangs spark cluster in yarn-client mode
> -------------------------------------------------------------
>
> Key: SPARK-20079
> URL: https://issues.apache.org/jira/browse/SPARK-20079
> Project: Spark
> Issue Type: Bug
> Components: YARN
> Affects Versions: 2.1.0
> Reporter: Guoqiang Li
>
> The ExecutorAllocationManager.reset method is called when re-registering AM,
> which sets the ExecutorAllocationManager.initializing field true. When this
> field is true, the Driver does not start a new executor from the AM request.
> The following two cases will cause the field to False
> 1. A executor idle for some time.
> 2. There are new stages to be submitted
> After the a stage was submitted, the AM was killed and restart ,the above two
> cases will not appear.
> 1. When AM is killed, the yarn will kill all running containers. All execuotr
> will be lost and no executor will be idle.
> 2. No surviving executor, resulting in the current stage will never be
> completed, DAG will not submit a new stage.
> Reproduction steps:
> 1. Start cluster
> {noformat}
> echo -e "sc.parallelize(1 to 2000).foreach(_ => Thread.sleep(1000))" |
> ./bin/spark-shell --master yarn-client --executor-cores 1 --conf
> spark.shuffle.service.enabled=true --conf
> spark.dynamicAllocation.enabled=true --conf
> spark.dynamicAllocation.maxExecutors=2
> {noformat}
> 2. Kill the AM process when a stage is scheduled.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
