hadoop-yarn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tsuyoshi Ozawa <oz...@apache.org>
Subject Re: Why is the NodeManager memory in the MiniYARNCluster hard-coded
Date Fri, 23 Jan 2015 05:52:07 GMT
Hi Robert,

Thank you for creating JIRA. Let's discuss it on the JIRA not to split
information between JIRA and mailing list.

Thanks,
- Tsuyoshi

On Fri, Jan 23, 2015 at 2:04 AM, Robert Metzger <rmetzger@apache.org> wrote:
> Hi,
> cool. I've filed a JIRA: https://issues.apache.org/jira/browse/YARN-3086
>
> If you want to maintain the current value of 4GB as the default value, I
> probably need to introduce a new configuration value, similar to:
>
> public static final String YARN_MINICLUSTER_CONTROL_RESOURCE_MONITORING =
>     YARN_PREFIX + "minicluster.control-resource-monitoring";
>
> Or is there another approach you would prefer?
>
> I'll add a patch to the JIRA once I've found time to fix it.
>
>
> Regards,
>
> Robert
>
>
>
>
> On Thu, Jan 22, 2015 at 8:15 AM, Tsuyoshi Ozawa <ozawa@apache.org> wrote:
>
>> Hi Robert,
>>
>> No, there is no limitation about this. We can make the configuration
>> configurable. I think we should make the default value 4GB for
>> backward compatibility. Could you create a JIRA for tracking the
>> issue?
>>
>> Thanks,
>> - Tsuyoshi
>>
>>
>> On Thu, Jan 22, 2015 at 1:27 AM, Robert Metzger <rmetzger@apache.org>
>> wrote:
>> > Hi,
>> >
>> > It seems that the memory per NodeManager is hardcoded in the
>> > MiniYARNCluster. I would like to pass a custom value to the
>> MiniYARNCluster.
>> >
>> >
>> https://github.com/apache/hadoop/blob/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-tests/src/test/java/org/apache/hadoop/yarn/server/MiniYARNCluster.java#L512
>> >
>> > Is there a reason for this limitation?
>> > Would you accept a patch which makes this value configurable?
>> >
>> >
>> > Best,
>> > Robert
>>

Mime
View raw message