hadoop-mapreduce-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vinod Kumar Vavilapalli <vino...@hortonworks.com>
Subject Re: Submitting jobs to classic MR using new client API
Date Tue, 04 Oct 2011 15:30:29 GMT
If you want to classic MR, you will need the classic jar generated by "ant
jar" on the classpath in which case JobTrackerClientProtocolProvider will be
used.

In any case, with hadoop 0.23 and trunk, classic MR is not going to be
supported. 0.20.* releases and (possibly) 0.22 comprise the last set of
releases that support classic MR with JT and TT.

HTH,
+Vinod

On Tue, Oct 4, 2011 at 7:57 PM, Kamesh <kamesh.b@imaginea.com> wrote:

> Hi,
>
> While debugging the YARN code, I come across following code snippet in
> YarnClientProtocolProvider.**java
>
>  @Override
>  public ClientProtocol create(Configuration conf) throws IOException {
>    if (MRConfig.YARN_FRAMEWORK_NAME.**equals(conf.get(MRConfig.**FRAMEWORK_NAME)))
> {
>      return new YARNRunner(conf);
>    }
>    return null;
>  }
>
>  @Override
>  public ClientProtocol create(InetSocketAddress addr, Configuration conf)
>      throws IOException {
>    return create(conf);
>  }
>
> By the second API, it is clear that we can connect to the classic MR using
> client API (using JobClient(jt address, conf) and hence submit the job using
> the same job client instance.).  However while initializing the cluster
> instance, the above functionality either creates a YarnRunner instance or
> returns null and hence client gets IOException. With this limitation we
> can't submit jobs to classic MR.
>
> There is any reason behind providing this?
>
> Please correct me if I understood wrongly.
>
> --
> Thanks&Regards,
> Bh.V.S.Kamesh.
>
>

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