Create a Yothalot program

After having implemented your mapreduce algorithm you have to create a little program that uses your implementation. The program should look as follows:

 *  myMapReduce.cpp
 *  main program that is passed to the Yothalot framework

 *  includes
#include <yothalot.h>
#include "MyMapReduce.h"

 *  the main program
int main(int argc, const char* argv[])
   // Instantiate your class
   MyMapReduce myMR;

   // Construct a Yothalot::Program object. You pass the instantiation of
   // your mapreduce class as a first argument and you pass argc and argv
   // as the second and third argument, so that the yothalot framework
   // can process the command line arguments to find out whether to start
   // the algorith in map, reduce or write state
   Yothalot::Program myYothalotProgram(argc, argv, &myMR);

   // run, run, run!

You first initialize the class with your mapreduce algorithm. Here we used a default initialization of a MapReduce class, but you can of course add other functionality to your class with other constructors as well that you use when you instantiate it.

Next, you create an instance of Yothalot::Program, by calling its constructor with arguments argc, argv and a pointer to the instance of your class. By passing argc, and argv to the constructor of Yothalot::Program, all command line arguments are passed to Yothalot::Program. This enables the Yothalot framework to call your program with the appropriate command line arguments to get the behavior of the program Yothalot wants. Passing a pointer to your algorithm is of course necessary to Yothalot::Program is aware of your algorithm.

Finally, you return with a call to member function run() of the Yothalot::Program class. That is all there is to create the executable.

Compiling your program

Since you are using C++ to implement your algorithm, you need to compile your program in order to use it. You may use your preferred compiler but we are at least sure that it works with gcc 4.8.4. You can install gcc using the repository of your Linux distribution.

If you are using gcc you can compile your program by typing on the command line in the directory where your program is stored:

g++ -std=c++11 myMapReduce.cpp -lyothalot -o myMapReduce

Since the Yothalot C++ API uses C++ 11 features you need to pass the -std=c++11 flag. Moreover, the program depends on the Yothalot library. You need to tell this to the compiler as well. This is done with -lyothalot.

Running the above command will give you a program myMapReduce. The Yothalot framework will call this program from its nodes. Therefore, the program should be available on all the nodes. If you copy your file to the GlusterFS cluster, you are sure that all nodes have access to the program.

You can of course add the optimization flags that you like, however, you have to make sure that if you allow the compiler to use certain CPU features these features should be supported by all CPUs in your Yothalot cluster.

Running your program

You are now ready to execute this MapReduce job on the Yothalot framework. Make sure Yothalot is running on the master node. If you are running locally this means that the Yothalot application should be started.

You can start a MapReduce job by calling Yothalot with the command line option --mapreduce, the name of your program, and some data. In case of the example MapReduce this yields:

./yothalot --mapreduce /path/to/myMapReduce word1 word2 word3 word1

After executing correctly, the output file on GlusterFS should contain three different word counts.

See Starting a Yothalot job for more detailed information on running jobs.

Access and paths

Yothalot needs to know where files and other resources are located on the GlusterFS file system. Therefore, it is important that you provide the absolute paths to these resources. To solve help you with this issue we have created a utility class Yothalot::Fullname that makes switching between relative and absolute paths easy. By using this class in your program you can pass either relative or absolute paths to your program and use absolute paths if you want to access a file or other resource.

Yet, there is one issue that Yothalot::Fullname does not solve, the path to your program. Your program is started up to perform mapper, reducer, and writer tasks. Therefore Yothalot needs to have access to your program. You can achieve this by putting your program on the Yothalot cluster. When you provide Yothalot the information about your program, you either have to provide the full path to your program, or you have add the program to the the $PATH environment. Then Yothalot knows where to look for your program.