[ https://issues.apache.org/jira/browse/SPARK-22319?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16214552#comment-16214552 ]
Apache Spark commented on SPARK-22319: -------------------------------------- User 'sjrand' has created a pull request for this issue: https://github.com/apache/spark/pull/19554 > SparkSubmit calls getFileStatus before calling loginUserFromKeytab > ------------------------------------------------------------------ > > Key: SPARK-22319 > URL: https://issues.apache.org/jira/browse/SPARK-22319 > Project: Spark > Issue Type: Bug > Components: Deploy, Spark Core > Affects Versions: 2.2.0, 2.3.0 > Reporter: Steven Rand > Assignee: Steven Rand > Fix For: 2.3.0 > > > In the SparkSubmit code, we call {{resolveGlobPaths}}, which eventually calls > {{getFileStatus}}, which for HDFS is an RPC call to the NameNode: > https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala#L346. > We do this before we call {{loginUserFromKeytab}}, which is further down in > the same method: > https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala#L655. > The result is that the call to {{resolveGlobPaths}} fails in secure clusters > with: > {code} > javax.security.sasl.SaslException: GSS initiate failed [Caused by > GSSException: No valid credentials provided (Mechanism level: Failed to find > any Kerberos tgt)] > {code} > A workaround is to {{kinit}} on the host before using spark-submit. However, > it's better if this workaround isn't necessary. A simple fix is to call > loginUserFromKeytab before attempting to interact with HDFS. > At least for cluster mode, this would appear to be a regression caused by > SPARK-21012. -- This message was sent by Atlassian JIRA (v6.4.14#64029) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org