[
https://issues.apache.org/jira/browse/MAPREDUCE-5329?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13748477#comment-13748477
]
Avner BenHanoch commented on MAPREDUCE-5329:
--------------------------------------------
Hi Siddharth,
I read some documentation about Yarn architecture and I got a better
understanding of your points. I am trying to suggest a new solution:
I see 4 issues in the current implementation of AuxiliaryServices / NodeManager
in Yarn:
# MAPREDUCE-5329: APPLICATION_INIT is never sent to AuxiliaryServices other
than the built in ShuffleHandler. This is in contrast to [the following
Yarn/NodeManager
documentation|http://hortonworks.com/blog/apache-hadoop-yarn-nodemanager/],
which says: _"Auxiliary services are notified when an application’s first
container starts on the node"_
# YARN-886: APPLICATION_STOP is inconsistent with APPLICATION_INIT
# New issue: We should consider shuffleToken to be specific to the shuffle
provider
# New issue: AM should support multiple AuxiliaryServices, each with a distinct
service port
for #1 & #2 we have already created JIRA issues. *I strongly suggest creating
distinct JIRA issues for #3 & #4 as well*. This will allow progressing in
parallel and for the users to benefit from the fixes independently (without
binding one fix to the other).
Last comment, regarding #3, I think that perhaps we should leave shuffleToken
to be general to all shuffle providers. This is for 2 reasons:
* AFAICS, shuffleToken is based on jobToken and user credentials; hence, it is
not specific to the provider but it is to the job&user.
* In the shuffle-consumer side, the token is not specific to the
shuffle-consumer, but it is part of the reduceTask; hence, it is general for
all shuffle-consumers.
Hence all shuffle services can use the same ShuffleToken without any problem.
Please let me know what do you think.
Thanks for your help,
Avner
> APPLICATION_INIT is never sent to AuxServices other than the builtin
> ShuffleHandler
> -----------------------------------------------------------------------------------
>
> Key: MAPREDUCE-5329
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5329
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mr-am
> Affects Versions: 2.0.4-alpha
> Reporter: Avner BenHanoch
>
> APPLICATION_INIT is never sent to AuxServices other than the built-in
> ShuffleHandler. This means that 3rd party ShuffleProvider(s) will not be
> able to function, because APPLICATION_INIT enables the AuxiliaryService to
> map jobId->userId. This is needed for properly finding the MOFs of a job per
> reducers' requests.
> NOTE: The built-in ShuffleHandler does get APPLICATION_INIT events due to
> hard-coded expression in hadoop code. The current TaskAttemptImpl.java code
> explicitly call: serviceData.put (ShuffleHandler.MAPREDUCE_SHUFFLE_SERVICEID,
> ...) and ignores any additional AuxiliaryService. As a result, only the
> built-in ShuffleHandler will get APPLICATION_INIT events. Any 3rd party
> AuxillaryService will never get APPLICATION_INIT events.
> I think a solution can be in one of two ways:
> 1. Change TaskAttemptImpl.java to loop on all Auxiliary Services and register
> each of them, by calling serviceData.put (…) in loop.
> 2. Change AuxServices.java similar to the fix in: MAPREDUCE-2668
> "APPLICATION_STOP is never sent to AuxServices". This means that in case the
> 'handle' method gets APPLICATION_INIT event it will demultiplex it to all Aux
> Services regardless of the value in event.getServiceID().
> I prefer the 2nd solution. I am welcoming any ideas. I can provide the
> needed patch for any option that people like.
> See [Pluggable Shuffle in Hadoop
> documentation|http://hadoop.apache.org/docs/current/hadoop-mapreduce-client/hadoop-mapreduce-client-core/PluggableShuffleAndPluggableSort.html]
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
