[ 
https://issues.apache.org/jira/browse/BIGTOP-1136?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13915426#comment-13915426
 ] 

jay vyas commented on BIGTOP-1136:
----------------------------------

Heres' the outline for the way to write this test i think.  Please comment 
accordingly and i can put the patch in .  

1) Define System properties with TWO users (tom and sally) who are authorized 
to run a hadoop job.
2) Run a job as "tom", which writes to toms directory, confirm SUCCESS 
(sh.getRet() == 0)
3) Run a job as "tom" which writes to sallys directory, confirm FAILURE 
(sh.getRet() != 0)

These users could be specified as configuration parameters to the unit test i 
guess. 

Does this sound good?  If so I can create a 
mapred/TestMultitenantMapreduce.groovy ITest pretty easily.  I already have 
something similar that i use, just need to groovify it.

> Testing Multitenancy and Multiuser applications
> -----------------------------------------------
>
>                 Key: BIGTOP-1136
>                 URL: https://issues.apache.org/jira/browse/BIGTOP-1136
>             Project: Bigtop
>          Issue Type: New Feature
>          Components: Tests
>            Reporter: jay vyas
>            Priority: Minor
>
> One of the items which is important for emerging, more flexible hadoop 
> deployments is multitenancy.  
> JIRA's such as this one: https://issues.apache.org/jira/browse/MAPREDUCE-5571
> Are very hard to test, and it would be extremely useful to many people 
> deploying non standard hadoop environments , if they could test that 
> multitenant and multiuser support was functional on their cluster. 
> This is sort of related to the stress tests jira 
> https://issues.apache.org/jira/browse/BIGTOP-1065, also recently created.
> I'm not really even sure this is possible in a fully automated / iTest sort 
> of fashion... so to start, any thoughts on how bigtop should test multi user, 
> multi tenant workloads?  One simple way to do it would be to launch 4 or 5 
> "calculate pi" jobs at one time, maybe from different users?  But how could 
> we "predict" off hand the usernames which will be available and appropriate 
> to submit jobs on a cluster, without the user specifying them?
> In any case, bigtop smokes is ideal for this kind of testing, because this 
> kind of testing is really only applicable to a cluster which is being used by 
> for a real deployment.  Other tests (i.e. pig smoke tests) are well handled 
> and covered by the individual ecosystem projects.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Reply via email to