[
https://issues.apache.org/jira/browse/SPARK-19936?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15907532#comment-15907532
]
Stan Teresen edited comment on SPARK-19936 at 3/13/17 2:53 PM:
---------------------------------------------------------------
The cluster has plenty of the resources (3 r3.large AWS instances) and I ran
only one example on the whole cluster, and usually example runs normally (at
the same time I noticed that in most of those successful cases only one
executor is engaged). The problem occurs when 2 executors happen to start. Why
would those timeouts in BlockManager (you can see them in the log files
1.stderr, 2.stderr) be happening on getting remote block from another executor
node?
was (Author: sutasu):
The cluster has plenty of the resource (3 r3.large AWS instances) and I ran
only one example on the whole cluster, and usually example runs normally (at
the same time I noticed that in most of those successful cases only one
executor is engaged). The problem occurs when 2 executors happen to start. Why
would those timeouts (you can see then in the log files 1.stderr, 2.stderr) be
happening?
> Page rank example takes long time to complete
> ---------------------------------------------
>
> Key: SPARK-19936
> URL: https://issues.apache.org/jira/browse/SPARK-19936
> Project: Spark
> Issue Type: Bug
> Components: Block Manager, Mesos
> Affects Versions: 2.1.0
> Environment: CentOS 7, Mesos 1.1.0
> Reporter: Stan Teresen
> Attachments: 1.stderr, 2.stderr, pr.out
>
>
> Sometimes Page Rank example takes very long time to finish on Mesos due to
> exceptions on fetching remote block in RetryingBlockFetcher on executor sides.
> As it is seen in the log files attached it took ~30 min for an example to
> complete in AWS 3 nodes environment (1 Mesos master and 2 agents).
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
