spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jongyoul Lee (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (SPARK-6286) Handle TASK_ERROR in TaskState
Date Thu, 12 Mar 2015 06:54:39 GMT

    [ https://issues.apache.org/jira/browse/SPARK-6286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14358180#comment-14358180
] 

Jongyoul Lee edited comment on SPARK-6286 at 3/12/15 6:54 AM:
--------------------------------------------------------------

[~srowen] I saw release note and a new protocol {{TASK_ERROR}} which avoids falling in infinite
loop by making wrong task description. And I agree with you because this is not compatible
with older versions of Mesos, and in case of spark, {{TASK_ERROR}} occurs rarely - personally,
none - because taskSchedulerImpl makes mesosTask mechanically. However, if you need to code
for {{TASK_ERROR}} to prepare future releases - actually no one wants to downgrade mesos without
any reason -, I think there are two options for dealing with {{TASK_ERROR}}. One is to treat
it as same as {{TASK_LOST}}, and another one is killing this spark application itself because
this might be a kind of system error. I think it's enough to choose the first option for now,
and we make this code compatible with older versions.


was (Author: jongyoul):
[~srowen] I saw release note and a new protocol {{TASK_ERROR}} which avoids falling in infinite
loop by making wrong task description. And I agree with you because this is not compatible
with older versions of Mesos, and in case of spark, {{TASK_ERROR}} occurs rarely - personally,
none - because taskSchedulerImpl makes mesosTask mechanically. However, if you need to code
for {{TASK_ERROR}} to prepare future releases - actually no one wants to downgrade mesos without
any reason -, I think there are two options for dealing with {{TASK_ERROR}. One is to treat
it as same as {{TASK_LOST}}, and another one is killing this spark application itself because
this might be a kind of system error. I think it's enough to choose the first option for now,
and we make this code compatible with older versions.

> Handle TASK_ERROR in TaskState
> ------------------------------
>
>                 Key: SPARK-6286
>                 URL: https://issues.apache.org/jira/browse/SPARK-6286
>             Project: Spark
>          Issue Type: Bug
>          Components: Spark Core
>            Reporter: Iulian Dragos
>            Priority: Minor
>              Labels: mesos
>
> Scala warning:
> {code}
> match may not be exhaustive. It would fail on the following input: TASK_ERROR
> {code}



--
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


Mime
View raw message