hadoop-mapreduce-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Arun C Murthy <...@hortonworks.com>
Subject Re: ApplicationMaster appId/appAttemptId/clusterTimestamp
Date Wed, 21 Sep 2011 00:36:57 GMT

On Sep 20, 2011, at 5:32 PM, Hitesh Shah wrote:

> Hi Chris
> 
> You are right. At the moment, you do need to pass all of them in to your app master for
it to register. I will file a jira to simplify this to end up with something along the lines
where the RM implicitly passes such information to the app master without the client explicitly
setting it ( the AM will need to consume the info for registration though). 
> 

Thanks Hitesh - we can use a @appattemptid@ or some such nomenclature.

> thanks
> -- Hitesh 
> 
> 
> On Sep 20, 2011, at 2:26 PM, Chris Riccomini wrote:
> 
>> Hey Guys,
>> 
>> Sorry to bug again, but I think it's best to get this stuff into the mail archives.
>> 
>> I want to confirm that I'm starting my ApplicationMaster in the "proper" way.
>> 
>> Right now, my client is setCommand'ing this:
>> 
>> "java -cp './package/*' kafka.yarn.ApplicationMaster " + streamerClass + " " + tasks
+ " "
>>       + "1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stdout "
>>       + "2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stderr"
>> 
>> It seems to me that my ApplicationMaster needs an appId, appAttemptId, and clusterTimestamp
in order to register with the ResourceManager. The only way that I see to do this is to pass
it (via the client) as the first three params to my kafka.yarn.ApplicationMaster. I recall
Arun mentioning that not all of this is required, but it seems to me (based on the API) that
it is.
>> 
>> Should I be putting the three RM params after the kafka.yarn.ApplicationMaster param
in my setCommand?
>> 
>> Thanks!
>> Chris
> 


Mime
View raw message