You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "liujianhui (JIRA)" <ji...@apache.org> on 2017/01/09 03:16:58 UTC
[jira] [Comment Edited] (SPARK-18806) driverwrapper and executor
doesn't exit when worker killed
[ https://issues.apache.org/jira/browse/SPARK-18806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15810519#comment-15810519 ]
liujianhui edited comment on SPARK-18806 at 1/9/17 3:15 AM:
------------------------------------------------------------
Hi, Sean Owen, I found the root cause, when start the driverwrapper or coarseGrainedExecutorBackend, the worker url is linke `DriverWrapper spark://Worker@10.0.52.25:32777`, the host is ip, but when a disconnectedProcessEvent occure, the remoteAddress is hostname:port, detail could be found at
{code}
val clientAddr = RpcAddress(addr.getHostString, addr.getPort)
nettyEnv.removeOutbox(clientAddr)
dispatcher.postToAll(RemoteProcessDisconnected(clientAddr))
{code}
but in class `WorkerWatcher`, whether the remoteAddress is Worker according to the follow codes
{code}
private def isWorker(address: RpcAddress) = expectedAddress == address
{code}
that will by false, as a result that the address's host is hostname instead of IP
was (Author: liujianhui):
Hi, Sean Owen, I found the root cause, when start the driverwrapper or coarseGrainedExecutorBackend, the worker url is linke `DriverWrapper spark://Worker@10.0.52.25:32777`, the host is ip, but when a disconnectedProcessEvent occure, the remoteAddress is hostname:port, detail could be found at
{code}
val clientAddr = RpcAddress(addr.getHostString, addr.getPort)
nettyEnv.removeOutbox(clientAddr)
dispatcher.postToAll(RemoteProcessDisconnected(clientAddr))
{code}
> driverwrapper and executor doesn't exit when worker killed
> ----------------------------------------------------------
>
> Key: SPARK-18806
> URL: https://issues.apache.org/jira/browse/SPARK-18806
> Project: Spark
> Issue Type: Bug
> Components: Deploy
> Affects Versions: 1.6.1
> Environment: java1.8
> Reporter: liujianhui
>
> submit an application with standlone-cluster mode, and then the master will launch executor and driverwrapper on worker. They are all start WorkerWatcher to watch the worker, as a result, when the worker killed by manual, the driverwrapper and executor sometimes will not exit.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org