hadoop-mapreduce-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Amogh Vasekar <am...@yahoo-inc.com>
Subject Re: Does Using MultipleTextOutputFormat Require the Deprecated API?
Date Tue, 15 Dec 2009 05:30:52 GMT
Yes. Also attached is an old thread I have kept handy with me. Hope this helps you.


Thanks,
Amogh


On 12/11/09 10:07 PM, "Geoffry Roberts" <geoffry.roberts@gmail.com> wrote:

Amogh,

I don't have experience with patches for hadoop.

I take it that I apply this patch using the linux patch utility.

I further assume, I need only apply the latest patch, which is 5.

Am I correct.

On Wed, Dec 9, 2009 at 7:30 AM, Amogh Vasekar <amogh@yahoo-inc.com> wrote:
http://issues.apache.org/jira/browse/MAPREDUCE-370

You'll  have to work around for now / try to apply patch.

Amogh



On 12/9/09 8:54 PM, "Geoffry Roberts" <geoffry.roberts@gmail.com <http://geoffry.roberts@gmail.com>
> wrote:

Aaron,

I am using 0.20.1 and I'm not finding org.apache.hadoop.mapreduce.
lib.output.MultipleOutputs.  I'm using the download page where the tar ball is dated from
Sep.09.

Sounds like I need to look at the code repository.


On Tue, Dec 8, 2009 at 1:39 PM, Aaron Kimball <aaron@cloudera.com <http://aaron@cloudera.com>
> wrote:
Geoffry,

There are two MultipleOutputs implementations; one for the new API, one for the old one.

The new API (org.apache.hadoop.mapreduce.lib.output.MultipleOutputs) does not have a getCollector()
method. This is intended to work with org.apache.hadoop.mapreduce.Mapper and its associated
Context object.

The old API implementation of MO (org.apache.hadoop.mapred.lib.MultipleOutputs) is intended
to work with org.apache.hadoop.mapred.Mapper, Reporter, and friends.

If you're going to use the new org.apache.hadoop.mapreduce-based code, you should not need
to import anything in the mapred package. That having been said -- I just realized that the
new-API-compatible MultipleOutputs implementation is not in Hadoop 0.20. It's only in the
unreleased 0.21. If you're using 0.20, you should probably stick with the old API for your
process.

Cheers,
- Aaron


On Tue, Dec 8, 2009 at 12:40 PM, Geoffry Roberts <geoffry.roberts@gmail.com <http://geoffry.roberts@gmail.com>
> wrote:
All,

This one has me stumped.

What I want to do is output from my reducer multiple files, one for each key value. I also
want to avoid any deprecated parts of the API.

As suggested, I switched from using MultipleTextOutputFormat to MultipleOutputs but have run
into an impasse.  MultipleOutputs' getCollector method requires a Reporter as a parameter,
but as far as I can tell, the API doesn't support this.  The only reporter I can find is in
the context object, but is declared protected.

Am I stuck? or just missing something?

My code:

@Override
public void reduce(Text key, Iterable<Text> values, Context context)
                throws IOException {
String fileName = key.toString();
             MultipleOutputs.addNamedOutput((JobConf) context.getConfiguration(), fileName,
OutputFormat.class, Text.class, Text.class);
            mos = new MultipleOutputs((JobConf) context.getConfiguration());
            for (Text line : values) {

// This is the problem line:
                mos.getCollector(fileName, <reporter goes here>).collect(
                        key, line);
            }

            mos.close();

         }

On Mon, Oct 5, 2009 at 11:17 AM, Aaron Kimball <aaron@cloudera.com <http://aaron@cloudera.com>
> wrote:
Geoffry,

The new API comes with a related OF, called MultipleOutputs (o.a.h.mapreduce.lib.output.MultipleOutputs).
You may want to look into using this instead.

- Aaron


On Tue, Sep 29, 2009 at 4:44 PM, Geoffry Roberts <geoffry.roberts@gmail.com <http://geoffry.roberts@gmail.com>
> wrote:
All,

What I want to do is output from my reducer multiple files one for each key value.

Can this still be done in the current API?

It seems that using MultipleTextOutputFormat requires one to use deprecated parts of API.

It this correct?

I would like to use the class or its equivalent and stay off anything deprecated.

Is there a work around?

In the current API one uses Job and a class derived from the class org.apache.hadoop.mapreduce.OutputFormat.
 MultipleTextOutputFormat does not derive from this class.

Job.setOutputFormatClass(Class<? extends org.apache.hadoop.mapreduce.OutputFormat>);


In the Old, deprecated API, one uses JobConf and an implementation of the interface org.apache.hadoop.mapred.OutputFormat.
 MultipleTextOutputFormat is just such an implementation.

JobConf.setOutputFormat(Class<? extends org.apache.hadoop.mapred .OutputFormat);








Mime
View raw message