[ 
https://issues.apache.org/jira/browse/SPARK-12646?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15092460#comment-15092460
 ] 

Marcelo Vanzin edited comment on SPARK-12646 at 1/13/16 7:36 PM:
-----------------------------------------------------------------

Can you convince people to at least use proper credentials to launch the Spark 
jobs instead of reusing YARN's?

I'm a little wary of adding this feature just to support a broken use case. 
When running on YARN, Spark is a user application, and you're asking for Spark 
to authenticate using service principals. That's kinda wrong, even if it works.

Your code also has a huge problem in that it uses {{InetAddress.getLocalHost}}; 
even if this were a desirable feature, there's no guarantee that's the correct 
host to use at all. On multi-homed machines, for example, which should be the 
address to use when expanding the principal template?

You application can also login to kerberos before launching the Spark job; call 
kinit by yourself and then launch Spark without using {{--principal}} nor 
{{--keytab}}. Then Spark doesn't need to do anything, it just inherits the 
kerberos ticket from your app.


was (Author: vanzin):
Can you convince people to at least use proper credentials to launch the Spark 
jobs instead of reusing YARN's?

I'm a little wary of adding this feature just to support a broken use case. 
When running on YARN, Spark is a user application, and you're asking for Spark 
to authenticate using service principals. That's kinda wrong, even if it works.

Your code also has a huge problem in that it uses {{InetAddress.getLocalHost}}; 
even if this were a desirable feature, there's no guarantee that's the correct 
host to use at all. On multi-homed machines, for example, which should be the 
address to use when expanding the principal template?

You application can also login to kerberos before launching the Spark job; call 
kinit by yourself and then launch Spark without using "--principal" nor 
"--keytab". Then Spark doesn't need to do anything, it just inherits the 
kerberos ticket from your app.

> Support _HOST in kerberos principal for connecting to secure cluster
> --------------------------------------------------------------------
>
>                 Key: SPARK-12646
>                 URL: https://issues.apache.org/jira/browse/SPARK-12646
>             Project: Spark
>          Issue Type: Improvement
>          Components: YARN
>            Reporter: Hari Krishna Dara
>            Priority: Minor
>              Labels: security
>
> Hadoop supports _HOST as a token that is dynamically replaced with the actual 
> hostname at the time the kerberos authentication is done. This is supported 
> in many hadoop stacks including YARN. When configuring Spark to connect to 
> secure cluster (e.g., yarn-cluster or yarn-client as master), it would be 
> natural to extend support for this token to Spark as well. 



--
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

Reply via email to