[ https://issues.apache.org/jira/browse/SPARK-4997?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14261285#comment-14261285 ]
Josh Rosen commented on SPARK-4997: ----------------------------------- I've noticed this issue on {{spark-ec2}} because that's generally the only test environment that I personally use, but it might be a more broad issue. In SPARK-2913 (the EC2 issue), the problem seems to be that the Hadoop directory is placed ahead of Spark on the classpath. My proposal was to just force Spark's file to appear at the head of the classpath, but I think that someone voiced a concern that this could break users' ability to use a custom log4j.properties file from their application JAR (or something like that). At a minimum, we should improve the documentation on the configuration page to discuss classpath ordering issues, since now it's just a fairly simplistic paragraph that just says "edit this file and it will work" and that's not the whole story (in fact, there are a bunch of StackOverflow questions from users who are confused about why the Spark log4j.properties file didn't take effect). > Check if Spark's conf needs to be put ahead of Hadoop's (for log4j purposes) > ---------------------------------------------------------------------------- > > Key: SPARK-4997 > URL: https://issues.apache.org/jira/browse/SPARK-4997 > Project: Spark > Issue Type: Bug > Components: EC2 > Affects Versions: 1.2.0 > Reporter: Nicholas Chammas > Priority: Minor > > As originally proposed in this PR: https://github.com/mesos/spark-ec2/pull/55 -- 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