[Hadoop-common-user] streaming job written in c++

For C/C++, you should be using the pipes interface.

Thank you Allen for you reply. I am trying to use MapReduce on Amazon EC2.
EC2 don't seem to support Pipes using their simple web GUI interface (is it
possible to use Pipes using their CLI or API?!). What is the problem with
using C++ with streaming?

Upendra

----- Original Message -----
From: "Allen Wittenauer" <awittenauer@linkedin.com>
To: <common-user@hadoop.apache.org>
Sent: Friday, December 04, 2009 6:17 PM
Subject: Re: streaming job written in c++

Absolutely none. You just read stdin/stdout as you would in any other
language. But since C/C++ has its own interface, it is just more common to
use that interface than using the one built for 'everything else'.

I've written plenty of apps in c++ using the streaming interface-- you
just need to read from std::cin and write to std::cout. Keys and
Values are separated by a tab.

Since Pipes doesn't give you access to very much of the Hadoop/MR
runtime (e.g., HDFS), and it's not a very idiomatic c++ interface
(e.g., it passes around vectors of strings by value) and doesn't
provide any of the usual JobConf convenience functions, I haven't
found reason to use it.

Chris

On Fri, Dec 4, 2009 at 6:17 PM, Allen Wittenauer
wrote: