[ 
https://issues.apache.org/jira/browse/MAPREDUCE-7044?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Jin Yibo updated MAPREDUCE-7044:
--------------------------------
    Description: When we use distributed scheduling at nodemanager, the 
amrmproxy will replace the token of app with the local one (we suppose 
amrmproxy will be always on when we need distributed scheduling). However, if 
we only enable the distributed scheduling in yarn.xml and disable the 
amrmproxy, the am will directly connect with rm instead of amrmproxy and then 
lead to invalid amrmtoken. Thus, we should add some checks to avoid distributed 
scheduling beyond non-amrmproxy (amrmproxy and distributed scheduling are two 
settings independently in yarn.xml).  (was: When we use distributed scheduling 
at nodemanager, the amrmproxy will replace the token of app with the local one 
(we suppose amrmproxy will be always on when we need distributed scheduling). 
However, if we enable the distributed scheduling in yarn.xml, the am will 
directly connect with rm instead of amrmproxy and then lead to invalid 
amrmtoken. This is because the token is replaced by amrmproxy. Therefore, we 
should tell am to consult amrmproxy at local nodemanager as its scheduler 
instead of rm directly when we want distributed scheduling.)

> Distributed Scheduling has invalid AMRMToken
> --------------------------------------------
>
>                 Key: MAPREDUCE-7044
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-7044
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mr-am
>    Affects Versions: 3.0.0-alpha2
>            Reporter: Jin Yibo
>            Priority: Minor
>             Fix For: 3.0.0-alpha2
>
>
> When we use distributed scheduling at nodemanager, the amrmproxy will replace 
> the token of app with the local one (we suppose amrmproxy will be always on 
> when we need distributed scheduling). However, if we only enable the 
> distributed scheduling in yarn.xml and disable the amrmproxy, the am will 
> directly connect with rm instead of amrmproxy and then lead to invalid 
> amrmtoken. Thus, we should add some checks to avoid distributed scheduling 
> beyond non-amrmproxy (amrmproxy and distributed scheduling are two settings 
> independently in yarn.xml).



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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

Reply via email to