A simple Hadoop with Z2 sample¶

This sample is an adaptation of the classical Wordcount sample in the Z2 context. This sample is supposed to show you how Hadoop can be used from within Z2 and in particular how to write Map/Reduce jobs in that context.

Note #1: This sample is made to be run on Linux or Mac-OS. Supposedly it is possible to run Hadoop on Windows. Sorry, but we have not been able to adapt the sample yet. A machine with 8GB of RAM should be sufficient.
Note #2: For your convenience everything in this sample assumes you use Eclipse. That as such is of course no prerequisite to running the software, but it just makes everything much more integrated for now. Please have Eclipse ready and the Eclipsoid installed. See How to install Eclipsoid.

This sample is provided by the repository z2-samples-hadoop-basic.

Prerequisites¶

This sample makes use of the Hadoop add-on that is based on Cloudera's CDH4 distribution of Hadoop. As client access is version dependent, so is the sample. In order to simplify this for you, there is a pre-configured CDH4 distribution available to you from this site. Apart from its development style configuration (i.e. no security), this is anyway the way we prefer to install Hadoop and friends: Just one root installation folder, one OS user, one log folder etc.

Please follow the procedure described here: Install prepacked CDH4.

To use with this sample, it is most convenient, if you clone and configure the CDH4 install next to your Eclipse workspace and the sample repository clone.

Setting up the sample¶

From here on, the sample is run like all samples, that is, following How to run a sample.

Assuming everything (including the CDH4 setup) is under install and your workspace in in install/workspace please clone z2-samples-hadoop-basic under install as well. Either from the command line as

cd install
git clone -b master http://git.z2-environment.net/z2-samples.hadoop-basic

or from within Eclipse using the Git repositories view (but make sure the folder is right next to your z2-base.core clone).

You should have an Eclipse workspace and next to it z2-samples.hadoop-basic, z2-samples.cdh4-base, and z2-base.core. Import all projects into your workspace.

We assume that you followed the steps in Install prepacked CDH4 and Hadoop is running (we do not need HBase in this case).

Running the sample¶

Start Z2.

This will take a short moment. When up, we want to first write a file into the Hadoop file system that we are going to split into words and count their occurences later on.

If you want to load some file you already have at hand, use the "copyFromLocal" operation to copy it into /hadoop-wordcount/input. E.g. if the file is called myfile.txt go into the CDH4 install and run

. ./env.sh 
hadoop fs -copyFromLocal myfile.txt /hadoop-wordcount/input

(the env.sh call is only required once per shell session).

Alternatively there is a z2Unit test (see How to z2Unit) that you can invoke to generate some input. As that is interesting on its own right, here is how that is done.

You should already have all the projects, in particular com.zfabrik.samples.hadoop-basic.wordcount already in your workspace. Otherwise import them from the repository you cloned previously.

You Eclipsoid to resolve all required compile dependency (Alt-R or click on the right Z in the toolbar), if you have not done so already.

Look for the type WriteWordsFile (Ctrl+Shift+T).

The method writeWordsFile will write a file of 100 million words in lines containing between 1 and 9 words each (but you can change that of course). Invoke it by right-clicking and "Run as / JUnit test". If you want to play around with the settings, simply change than, synchronize z2 (Alt-Y or click on the left Z in the toolbar) and rerun.

The interesting piece about this code is how it is connecting to HFDS:

...
    @Test
    public void writeWordsFile() throws Exception {
        FileSystem fs = FileSystem.get(IComponentsLookup.INSTANCE.lookup(WordCountMRJob.CONFIG, Configuration.class));
        fs.delete(WordCountMRJob.INPUT_PATH, true);
        fs.mkdirs(WordCountMRJob.INPUT_PATH.getParent());
...

Here, the actual connection configuration, one of Hadoop's XML configuration files, is looked up from a Z2 component called com.zfabrik.samples.hadoop-basic.wordcount/nosql. The component type for that is defined by the Hadoop integration module "com.zfabrik.hadoop" of the Hadoop add on.

The purpose of this is to separate the client configuration information from the using implementation. We will see another application of that below.

So now we assume you have the input file uploaded or generated in HDFS.