[ https://issues.apache.org/jira/browse/SPARK-5164?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14270533#comment-14270533 ]
Aniket Bhatnagar commented on SPARK-5164: ----------------------------------------- First issue can be fixed by using Environment.<variable>.$$() instead of Environment.<variable>.$() in ClientBase. But unfortunately, $$() method seems to be only added in recent versions of hadoop making it not a viable option if we want to support many versions of Hadoop. I am not sure if it is possible to detect remote OS using YARN API. I am thinking that perhaps we should introduce a new configuration - spark.yarn.remote.os that hints about the target YARN OS an can take values - Windows or Linux. We can then use this configuration in ClientBase and Path.SEPARATOR. I am happy to submit a pull request for this, once the recommendation is vetted by the community. > YARN | Spark job submits from windows machine to a linux YARN cluster fail > -------------------------------------------------------------------------- > > Key: SPARK-5164 > URL: https://issues.apache.org/jira/browse/SPARK-5164 > Project: Spark > Issue Type: Bug > Components: YARN > Affects Versions: 1.2.0 > Environment: Spark submit from Windows 7 > YARN cluster on CentOS 6.5 > Reporter: Aniket Bhatnagar > > While submitting spark jobs from a windows machine to a linux YARN cluster, > the jobs fail because of the following reasons: > 1. Commands and classpath contain environment variables (like JAVA_HOME, PWD, > etc) but are added as per windows's syntax (%JAVA_HOME%, %PWD%, etc) instead > of linux's syntax ($JAVA_HOME, $PWD, etc). > 2. Paths in launch environment are delimited by semi-colon instead of colon. > This is because of usage of Path.SEPARATOR in ClientBase and > YarnSparkHadoopUtil. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org