hadoop-yarn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sergiy Matusevych <sergiy.matusev...@gmail.com>
Subject Re: YARN Client and Unmanaged AM running in the same process?
Date Tue, 24 Jan 2017 19:28:58 GMT
Hi Jason,

Thanks a lot for your help! It is great to hear that Unmanaged AM mode
works at least on some version of YARN :) I will check our YARN logs and
give you more details soon.

Thank you!
Sergiy.

On Tue, Jan 24, 2017 at 8:48 AM, Jason Lowe <jlowe@yahoo-inc.com> wrote:

> Have you looked at the ResourceManager logs to see what it is doing when
> it receives the unregister request?  I'm wondering if there's an exception
> or error that could explain why it is not working as expected.  The sample
> code works against trunk for me.  The unmanaged AM registered,
> unregistered, and the final status of the application on the RM was
> FINISHED/SUCCEEDED.
>
> Jason
>
>
> On Monday, January 23, 2017 9:51 PM, Sergiy Matusevych <
> sergiy.matusevych@gmail.com> wrote:
>
>
> Hi fellow YARN developers,
>
> I am writing a YARN application that runs both Client *and* Unmanaged
> Application Master in the same JVM process. I have a toy example that
> starts a YARN application and the AM in Unmanaged mode, and then just shuts
> it down:
>
> https://github.com/apache/reef/blob/master/lang/java/
> reef-runtime-yarn/src/test/java/org/apache/reef/runtime/
> yarn/driver/unmanaged/UnmanagedAmTest.java
>
> (I wrapped it as a unit test, but the code is 100% independent of REEF, so
> you can copy & paste it if you want to play with it; I can also build a
> small maven project around it).
>
> The app *almost* works - the problem seems to be that the call on line 117
>
>     rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED,
> "Success!", null);
>
> fails to update the status of the application on the Resource Manager side.
> I would expect it to be FINISHED/SUCCEDED, but instead it stays in
> RUNNING/UNDEFINED well after the client/AM process terminates, and
> eventually got marked as FAILED/FAILED by the RM.
>
> Am I doing something wrong, or is that a bug in YARN? I found an old JIRA
> issue that seems to be related to the problem:
>
> https://issues.apache.org/jira/browse/YARN-273
> "Add an unmanaged AM client for in-process AMs"
>
> Can someone confirm if my problem is indeed related to that issue, or is
> there something wrong with my code?
>
> Thank you!
> Sergiy.
>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message