[
https://issues.apache.org/jira/browse/SPARK-7736?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14620021#comment-14620021
]
Esben S. Nielsen commented on SPARK-7736:
-----------------------------------------
Thanks for the comment. I don't understand how it apply here however as both
listed pyspark programs (In my understanding) should result in step 2) of your
scenario:
p1) Unhandled execption raised before SparkContext initialization:
---
from pyspark import SparkContext
raise Exception('Fail')
sc = SparkContext(appName="raise_seen_by_yarn")
---
This results in an AM retry (total 2 AM tries as per YARN default) and
subsequent marking of the application YARN status as FAILED. This is what I
expect for a "designed to fail AM".
p2) Unhandled execption raised after SparkContext initialization:
---
from pyspark import SparkContext
sc = SparkContext(appName="raise_not_seen_by_yarn"):
raise Exception('Fail')
---
This results in the the application being marked as SUCCEEDED (total of 1 AM
try) which is not what I expect for a "designed to fail AM".
I've tried to look in the spark documentation if there should be taken special
actions to signal failure to YARN but I haven't found anything? And looking at
src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala : L118
where all sys.exit calls are considered successful termination regardless of
exit code I can't see a way to signal failure to YARN after SparkContext
initialization?
Both p1 and p2 return with non-zero exit code when run with spark-submit
--master yarn-client which is what I would expect.
> Exception not failing Python applications (in yarn cluster mode)
> ----------------------------------------------------------------
>
> Key: SPARK-7736
> URL: https://issues.apache.org/jira/browse/SPARK-7736
> Project: Spark
> Issue Type: Bug
> Components: YARN
> Environment: Spark 1.3.1, Yarn 2.7.0, Ubuntu 14.04
> Reporter: Shay Rojansky
>
> It seems that exceptions thrown in Python spark apps after the SparkContext
> is instantiated don't cause the application to fail, at least in Yarn: the
> application is marked as SUCCEEDED.
> Note that any exception right before the SparkContext correctly places the
> application in FAILED state.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
