How to Debug BSP Programs

Debugging distributed programs is always difficult, because very few debuggers will let you connect to a remote program that wasn't run with the proper command line arguments.

1. Start by getting everything running (likely on a small input) in the local runner.

• You do this by setting your BSP Master to "local" in your config. The local runner can run under the debugger and runs on your development machine. A very quick and easy way to set this

  config variable is to include the following line just before you run the job:
  conf.set("bsp.master.address", "local");
  
  Running in local mode makes the job run within 20 threads by default. Since this isn't always very convenient to debug, you can decrease the number of tasks with this line:
  conf.set("bsp.local.tasks.maximum", "2")
  Obviously, this sets the number of tasks used to 2.
  
  You may also want to do this to make the input and output files be in the local file system rather than in the Hadoop distributed file system (HDFS):
  conf.set("fs.default.name", "local");
  
  You can also set these configuration parameters in hama-site.xml. The configuration files should appear somewhere in your program's class path when the program runs.

2. Run the small input on a 1 node cluster. This will smoke out all of the issues that happen with

• distribution and the "real" task runner, but you only have a single place to look at logs. Besides the task logs, the most useful ones are the grooms and bspmaster logs. Make sure you are logging at the INFO level or you will miss clues like the output of your tasks.