Hi Ahmed,
Hadoop 0.20.0 included the new mapred API, these sometimes
refer as context objects. These are designed to make API easier to evolve in
future. There are some differences between new & old API's,
> The new API's favour abstract classes rather than interfaces, since abstract
> classes are easy to evolve.
> New API's use context objects like MapContext & ReduceContext to connect the
> user code.
> The old API has a special JobConf object for jobconf, in new API Job
> configuration will be done using Configuration.
You can find the new API's in org.apache.hadoop.mapreduce.lib.input.* package
and its sub packages, old API's in org.apache.hadoop.mapred.* package its sub
packages.
The new API is type-incompatible with the old, we need to rewrite the jobs to
make use of these advantages.
Based on these things you can select which API's to use.
Thanks
Devaraj k
From: Ahmed Eldawy [mailto:aseld...@gmail.com]
Sent: 05 July 2013 00:00
To: user@hadoop.apache.org
Subject: Which InputFormat to use?
Hi I'm developing a new set of InputFormats that are used for a project I'm
doing. I found that there are two ways to create a new InputFormat.
1- Extend the abstract class org.apache.hadoop.mapreduce.InputFormat
2- Implement the interface org.apache.hadoop.mapred.InputFormat
I don't know why there are two versions which are incompatible. I found out
that for each one, there is a whole set of interfaces for different classes
such as InputSplit, RecordReader and MapReduce job. Unfortunately, each set of
classes is not compatible with the other one. This means that I have to choose
one of the interfaces and go with it till the end. I have two questions
basically.
1- Which of these two interfaces I should go with? I didn't find any
deprecation in one of them so they both seem legitimate. Is there any plan to
retire one of them?
2- I already have some classes implemented in one of the formats, does it worth
refactoring these classes to use the other interface, in case I used he old
format.
Thanks in advance for your help.
Best regards,
Ahmed Eldawy
