[ 
https://issues.apache.org/jira/browse/OOZIE-2983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16110783#comment-16110783
 ] 

Andras Piros commented on OOZIE-2983:
-------------------------------------

[~rkanter] +1 on the original approach, your patch seems pretty advanced (apart 
from TODOs). I like the idea of Oozie being able to stream the launcher AM logs 
on its own, rather than relying on a running JobHistoryServer and the client's 
capability of following redirects in the first place.

I'd be curious how Spark does that, though. [~wilfreds] can you point to the 
Spark sources where Spark does the redirect? I [might be 
missing|https://github.com/apache/spark/blob/6735433cde44b3430fb44edfff58ef8149c66c13/resource-managers/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala#L376-L380]
 the place.

> Stream the Launcher AM Logs
> ---------------------------
>
>                 Key: OOZIE-2983
>                 URL: https://issues.apache.org/jira/browse/OOZIE-2983
>             Project: Oozie
>          Issue Type: Sub-task
>    Affects Versions: 5.0.0
>            Reporter: Robert Kanter
>            Assignee: Andras Piros
>         Attachments: OOZIE-2983.001.patch
>
>
> Because OYA will get rid of the MR-based Launcher Job and have a custom AM 
> instead, the Launcher Job info and logs will no longer show up in the JHS. It 
> will show up in the ATSv2, however, the ATSv2 isn't quite ready yet and that 
> doesn't help users on older versions of Hadoop.  Launcher Jobs will stil show 
> up in the RM, but the RM doesn't remember things forever (and can be quite 
> short depending on cluster busyness and configs).
> We should have Oozie serve those logs itself (which would also be a nice 
> thing to have anyway, as the launcher logs are typically the most useful). 
> The code for reading those files is pretty easy to call. The yarn CLI, JHS, 
> and this [Aggregated Log Files 
> Exploder|https://github.com/rkanter/ALFExploder] tool I wrote all use it; the 
> latter proves that you can do it outside of Hadoop. 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to