[ 
https://issues.apache.org/jira/browse/SPARK-16298?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15366310#comment-15366310
 ] 

Thomas Graves commented on SPARK-16298:
---------------------------------------

kinit as the user in the keytab before submitting the job.  
I remember there were a few issues with this originally and I'm not sure if 
they were all fixed. I thought if you weren't kinited yet it would use the 
keytab to do so, but I'm not sure it supposed changing the user.  

https://github.com/apache/spark/pull/9875  might be related. 

If we change it we should make it consistent and always just login from keytab 
or don't.

> spark.yarn.principal not working
> --------------------------------
>
>                 Key: SPARK-16298
>                 URL: https://issues.apache.org/jira/browse/SPARK-16298
>             Project: Spark
>          Issue Type: Bug
>            Reporter: Partha Pratim Ghosh
>
> I am opening a Spark configuration with spark.yarn.principal and 
> spark.yarn.keytab. However, this is not authenticating the underlying HDFS 
> with the same principal and keytab. Instead, seems it is picking up from 
> ticket cache. Without this feature the spark.yarn.principal and 
> spark.yarn.keytab doesn't seem to be logical.
> Sample code - 
> SparkConf conf = new 
> SparkConf().setMaster("yarn-client").setAppName("spark-test")
>                                               .set("spark.repl.class.uri", 
> classServerUri);
>                               conf.set("spark.yarn.principal", principal);
>                               conf.set("spark.yarn.keytab", keytab);
>                               conf.setSparkHome(sparkBasePath);
>                               
>                               if (execUri != null) {
>                                       conf.set("spark.executor.uri", execUri);
>                               }
>                               conf.set("spark.executor.memory", "8g");
>                               conf.set("spark.scheduler.mode", "FAIR");
>                               SparkContext sparkContext = new 
> SparkContext(conf);
> Please advise how this can be achieved.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to