Dumbo - Python module that allows you to easily write and run Hadoop programs | Hacker News Reader

Dumbo - Python module that allows you to easily write and run Hadoop programs

github.com

39 pointscoderdude13 comments

Open articleView on HN

At one point we used Dumbo for a lot of stuff at Yelp, but we finally ended up writing our own framework:

http://engineeringblog.yelp.com/2010/10/mrjob-distributed-co...

It's open source and easy to use, and makes hadoop on EMR exceptionally clean. Check it out.

Nice work on mrjob! We've been really happy with it.

And for anyone who didn't see it earlier: http://yahoo.github.com/oozie/

These four are the only one's I know of. Does anyone know of any others?

Unlike Dumbo, mrjob & hadoopy, Oozie isn't a Python Streaming library, so I'm not sure what you're looking for by 'others'?

I didn't mean to imply that it's another Python Streaming library. I included it because it's a library for working with Hadoop jobs. Unless I am mistaken in that? I haven't yet tried Oozie myself.

Right, it's just that there's a whole slew if you decide to leave the Python Streaming world (which I figured people reading this would care about).

Cascading, cascalog, wukong, innumerable others.

You were certainly justified in questioning what I was talking about. I didn't know there were so many others actually. Thanks for the names of a few, though.

http://discoproject.org/

Thats another project of similar vein. Although it has native support for Python and you dont need to install Hadoop (which may be better for some).

To be clear, Disco is a full replacement for Hadoop (including MapReduce, HDFS, etc). I've heard good things, but you also leave the (rather large) Hadoop ecosystem.

As a Python developer I've found Dumbo / mrjob to be easy to use.

There is that.

For those curious Disco has an Erlang/Python base. It does seem to be getting better too. I also found it easier to set up a Map/Reduce cluster in it then Hadoop.

Which do you prefer using, Dumbo or mrjob?

They're pretty similar, it's hard to make map reduce very complex in Python.

What mrjob has going for it is that it's very easy to run a job on Amazon's EMR (Elastic Map Reduce) so you don't have to do any Hadoop setup. You can run Dumbo on EMR, too, it's just more manual (last I checked).

That said, Dumbo has been around longer and seems to be smarter about input/output formats. It seems that by default it can ready just about any type of Hadoop format (look up SequenceFiles, for example) and by default it'll output compressed SequenceFiles, too, which help save space and I/O.

Reply on news.ycombinator.com