samza-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Bharath Kumara Subramanian <codin.mart...@gmail.com>
Subject Re: Running Samza with YARN Node label support
Date Wed, 18 Dec 2019 22:34:29 GMT
Hi Debraj,

I forgot to call this out earlier. Some distribution of YARN doesn't
support node label and rack combination as part of the same request. If you
were to use node labels along with host affinity feature
<https://samza.apache.org/learn/documentation/latest/yarn/yarn-host-affinity.html>
in Samza, you might run into following issue

19:25:10.032 [main] ClusterBasedJobCoordinator [ERROR] Exception thrown in
> the JobCoordinator loop
> org.apache.hadoop.yarn.client.api.InvalidContainerRequestException: Cannot
> specify node label with rack and node at
> org.apache.hadoop.yarn.client.api.impl.AMRMClientImpl.checkNodeLabelExpression(AMRMClientImpl.java:617)
> at


Refer https://jira.apache.org/jira/browse/YARN-4925
<https://jira.apache.org/jira/browse/YARN-4925?attachmentOrder=asc> for
more information. You may want to back-port the patch to your custom YARN
distribution if applicable.

Thanks,
Bharath

On Wed, Dec 18, 2019 at 1:15 PM Bharath Kumara Subramanian <
codin.martial@gmail.com> wrote:

> Hi Debraj,
>
> To get the node label working, set the label configurations[1] pointed out
> by Yang in your application config. Samza will take care of embedding the
> node label in the resource request automatically if it notices the label
> configuration inside your application.
> Samza framework respects node label configurations even though they are
> documented in the configuration table. I have created SAMZA-2422
> <https://issues.apache.org/jira/browse/SAMZA-2422> to track this work
> item.
>
> Let us know if you run into any issues.
>
> Thanks,
> Bharath
>
> [1] -
> *yarn.container.label* for specifying node label for the containers
> *yarn.am.container.label*  for specifying node label for the application
> master
>
> On Wed, Dec 18, 2019 at 10:49 AM Debraj Manna <subharaj.manna@gmail.com>
> wrote:
>
>> I understood how I can assign labels to yarn nodes.
>>
>> But it is still not clear to me how can I specify the node label for a
>> samza application. I am referring to the section "Specifying node label
>> for
>> application" in the link
>> <
>> https://hadoop.apache.org/docs/r2.7.3/hadoop-yarn/hadoop-yarn-site/NodeLabel.html
>> >
>> you shared in your last email.
>>
>> On Wed, Dec 18, 2019 at 11:17 PM Yang Zhang <zhan3248@umn.edu> wrote:
>>
>> > Hi Debraj Manna,
>> >
>> > The app-def in previous email is just an example where you can configure
>> > node labels. Yarn node labels
>> > <
>> >
>> https://hadoop.apache.org/docs/r2.7.3/hadoop-yarn/hadoop-yarn-site/NodeLabel.html
>> > >
>> > is
>> > a general feature (not specific to Samza), and it depends on the
>> > configuration system your system uses. The example uses xml format to
>> > configure Samza job, but Samza as a framework, it does not restrict
>> > configuration format. Please let us know if you have further questions,
>> and
>> > we should detail the documents in OSS to describe the usage of certain
>> > features.
>> >
>> > Best,
>> > Yang
>> >
>> > On Tue, Dec 17, 2019 at 9:58 PM Debraj Manna <subharaj.manna@gmail.com>
>> > wrote:
>> >
>> > > Thanks, Yang for replaying.
>> > >
>> > > Yes, my use case is almost similar.
>> > >
>> > > Can you let me know which app-def you are referring to? I am not able
>> to
>> > > locate yarn.am.container.label in samza-configurations
>> > > <
>> > >
>> >
>> http://samza.apache.org/learn/documentation/latest/jobs/samza-configurations.html
>> > > >
>> > > .
>> > > Is there any samza project whose code I can refer to regarding the
>> usage
>> > of
>> > > these configurations?
>> > >
>> > > On Wed, Dec 18, 2019 at 7:42 AM Yang Zhang <zhan3248@umn.edu> wrote:
>> > >
>> > > > Hello Debraj,
>> > > >
>> > > > We do not have a formal documentation in open source to describe how
>> > yarn
>> > > > node label is used in general. In contrast, we have an example of
>> using
>> > > > yarn node label to specify Samza container to run over "HDD" rather
>> > than
>> > > > default "SSD" nodes. Please take a look at the following guide and
>> let
>> > us
>> > > > know whether it can be applied for your use case. Thank you for
>> > reporting
>> > > > this issue!
>> > > > =================================================Step-by-step guide
>> > > >
>> > > >
>> > > >    1.
>> > > >
>> > > >    Add the *yarn.container.label *and* yarn.am.container.label* to
>> the
>> > > >    job's *app-def* if not already present. The default of an empty
>> > string
>> > > >    will keep the current default behavior of using SSD nodes.
>> > > >    <?xml version="1.0" encoding="UTF-8"?>
>> > > >    <application
>> > xmlns="urn:com:linkedin:ns:configuration:definition:1.0"
>> > > >    name="my-application" version="">
>> > > >        <configuration-definition>
>> > > >            <property name="yarn.am.container.label" default=""
/>
>> <!--
>> > > the
>> > > >    label used for launching the application master -->
>> > > >            <property name="yarn.container.label" default="" />
<!--
>> the
>> > > >    label used for other containers -->
>> > > >        </configuration-definition>
>> > > >    </application>
>> > > >    2. If you had to modify your *app-def* in step 1, you will need
>> to
>> > do
>> > > a
>> > > >    trigger-build to get the change to take effect.
>> > > >    3.
>> > > >
>> > > >    Add the label to *application.src* for your job. The *hdd* label
>> > will
>> > > >    assign your containers to machines with spinning disks instead
of
>> > > solid
>> > > >    state disks.
>> > > >    <?xml version="1.0" encoding="UTF-8"?>
>> > > >    <application xmlns="urn:com:linkedin:ns:configuration:source:1.0"
>> > > >    name="my-application">
>> > > >      <configuration-source>
>> > > >        <property name="yarn.container.label" value="hdd" />
>> > > >        <property name="yarn.am.container.label" value="hdd" />
>> > > >      </configuration-source>
>> > > >    </application>
>> > > >
>> > > >
>> > > >    4.
>> > > >
>> > > >    Deploy.
>> > > >
>> > > > =================================================
>> > > >
>> > > >
>> > > > Best,
>> > > >
>> > > > Yang
>> > > >
>> > > > On Tue, Dec 17, 2019 at 10:13 AM Debraj Manna <
>> > subharaj.manna@gmail.com>
>> > > > wrote:
>> > > >
>> > > > > Hi
>> > > > >
>> > > > > I am seeing running samza with yarn node label is resolved in
>> 0.12.
>> > > > >
>> > > > >
>> > > > >
>> > > >
>> > >
>> >
>> https://issues.apache.org/jira/browse/SAMZA-1013?page=com.atlassian.jira.plugin.system.issuetabpanels%3Aall-tabpanel
>> > > > >
>> > > > > But I am not able to locate the relevant documentation in
>> samza-yarn
>> > > > > documentation
>> > > > > <
>> > > >
>> >
>> https://samza.apache.org/learn/documentation/latest/deployment/yarn.html
>> > > >
>> > > > >
>> > > > > Can someone point me to the relevant documentation?
>> > > > >
>> > > >
>> > >
>> >
>>
>

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