[jira] [Commented] (SPARK-3289) Avoid job failures due to rescheduling of failing tasks on buggy machines

["Silvio Bernardinello (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-3289?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14620247#comment-14620247 ] 

Silvio Bernardinello commented on SPARK-3289:
---------------------------------------------

We are experimenting this issue running Spark 1.4 on Mesos cluster.
When a Mesos slave fails and there is a Spark executor running tasks in it, tasks are rescheduled on the same executor, which is no more reachable.

Mesos LOST tasks are not rescheduled by the Spark framework (see https://mail-archives.apache.org/mod_mbox/mesos-user/201310.mbox/%3CCAAkWvAxPRRNRCdLAZcybnmk1_9eLyhEOdAf8urf8ssrLBAcx8g@mail.gmail.com%3E).



> Avoid job failures due to rescheduling of failing tasks on buggy machines
> -------------------------------------------------------------------------
>
>                 Key: SPARK-3289
>                 URL: https://issues.apache.org/jira/browse/SPARK-3289
>             Project: Spark
>          Issue Type: Bug
>          Components: Spark Core
>            Reporter: Josh Rosen
>
> Some users have reported issues where a task fails due to an environment / configuration issue on some machine, then the task is reattempted _on that same buggy machine_ until the entire job failures because that single task has failed too many times.
> To guard against this, maybe we should add some randomization in how we reschedule failed tasks.



--
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