-RE: How to understand Hadoop source code ?
Noelle Jakusz 2013-04-18, 17:32
There are quite a few new people, so maybe start a collaborative group where you can collect notes and steps (videos and articles). I know I would have some for you that I have created as I have gotten started... it would be a great idea to post them after some collaboration and review.
Thanks Chris for the detailed reply...
From: Chris Nauroth [mailto:[EMAIL PROTECTED]]
Sent: Thursday, April 18, 2013 1:14 PM
To: [EMAIL PROTECTED]
Subject: Re: How to understand Hadoop source code ?
Is there a specific bug fix or feature that you are trying to contribute?
Specific questions like "how can I help with jira X?" or "what is the main entry point when I run the hdfs command?" or "where does the namenode serialize metadata to disk" or "where does the secondary namenode execute a checkpoint" can help focus the conversation.
AFAIK, we don't have a general code walkthrough document focused on onboarding new engineers. This could be a valuable contribution if you want to gather notes while you learn. I think this always works best if it's driven by a new engineer with review by an expert. (If the experts write it, then they might accidentally skip something non-obvious that they've already internalized.)
Since that document doesn't exist yet, the other option is to do some reading of the code, ideally while trying to fix a specific bug that has been filed in jira. Like you said, it's a relatively large codebase, so it's impractical to read the whole thing top-to-bottom. Instead, it's important to look for high-level clues that steer you towards the right files. I've found that the Maven module structure and the Java package names are usually descriptive enough to steer me in the right direction.
If you focus on getting familiar with those, you'll basically build a btree inside your brain that helps you index into the right part of the codebase and answer your own questions rapidly. Several examples:
"Where is the main entry point for the datanode daemon?": module hadoop-hdfs, package org.apache.hadoop.hdfs.server.datanode
"What is the algorithm for rebalancing an unbalanced cluster?": module hadoop-hdfs, package org.apache.hadoop.hdfs.server.balancer
"How does YARN launch a new container process?": module hadoop-yarn-server-nodemanager, package org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher
"Multiple daemons publish JMX metrics as a common concern. Where is that
implemented?": module hadoop-common, package org.apache.hadoop.metrics2
I hope this is helpful to get the process started for you. We're always here to help if you have specific follow-up questions.
On Wed, Apr 17, 2013 at 10:33 PM, Prabakaran Krishnan < [EMAIL PROTECTED]> wrote:
> Couuld you please help me in understand map reduce in Hadoop?
> From: Mohammad Mustaqeem <[EMAIL PROTECTED]>
> To: common-dev <[EMAIL PROTECTED]>
> Sent: Thursday, 18 April 2013 10:44 AM
> Subject: Re: How to understand Hadoop source code ?
> I am interested in HDFS. Please guide me.
> On Thu, Apr 18, 2013 at 3:36 AM, Arun C Murthy <[EMAIL PROTECTED]>
> > Please don't cross post.
> > What parts of Hadoop are you interested in? HDFS? YARN? MapReduce?
> > Arun
> > On Apr 17, 2013, at 2:50 PM, Mohammad Mustaqeem wrote:
> > > Hello everyone,
> > > I am new to this group. Since the source code of Hadoop
> > > is
> > > big, I am not able to understand it entirely.
> > > Is there any document that describes the code?
> > > Is there any way to understand the functionality of each classes
> > > and
> > > method?
> > >
> > >
> > > --
> > > *With regards ---*
> > > *Mohammad Mustaqeem*,
> > > M.Tech (CSE)
> > > MNNIT Allahabad
> > --
> > Arun C. Murthy
> > Hortonworks Inc.
> > http://hortonworks.com/
> *With regards ---*
> *Mohammad Mustaqeem*,
> M.Tech (CSE)