5.1. Simple Bag-of-Tasks¶
You might be wondering how to create your own RADICAL-Pilot script or how RADICAL-Pilot can be useful for your needs. Before delving into the remote job and data submission capabilities that RADICAL-Pilot has, its important to understand the basics.
The simplest usage of a pilot-job system is to submit multiple identical tasks (a ‘Bag of Tasks’ / ‘BoT’) collectively, i.e. as one big job! Such usage arises for example to perform parameter sweep jobs, or to execute ensemble simulation.
We will create an example which submits N tasks using RADICAL-Pilot. The tasks are all identical, except that they each record their number in their output. This type of run is very useful if you are running many jobs using the same executable (but perhaps with different input files). Rather than submit each job individually to the queuing system and then wait for every job to become active individually, you submit just one container job (called the Pilot). When this pilot becomes active, it executes your tasks on the reserved cores. RADICAL-Pilot names such tasks ‘Compute Units’, or short ‘CUs’.
Download the file
simple_bot.py with the following command:
curl -O https://raw.githubusercontent.com/radical-cybertools/radical.pilot/master/examples/docs/simple_bot.py
Open the file
simple_bot.py with your favorite editor. The example should
work right out of the box on your local machine. However, if you want to try it
out with different resources, like remote HPC clusters, look for the sections
# ----- CHANGE THIS -- CHANGE THIS -- CHANGE THIS -- CHANGE THIS ------
and change the code below accordging to the instructions in the comments.
This assumes you have installed RADICAL-Pilot either globally or in a Python virtualenv. You also need access to a MongoDB server.
Set the RADICAL_PILOT_DBURL environment variable in your shell to the MongoDB server you want to use, for example:
If RADICAL-Pilot is installed and the MongoDB URL is set, you should be good to run your program (the database is created on the fly):
The output should look something like this:
Initializing Pilot Manager ... Submitting Compute Pilot to Pilot Manager ... Initializing Unit Manager ... Registering Compute Pilot with Unit Manager ... Submit Compute Units to Unit Manager ... Waiting for CUs to complete ... ... Waiting for CUs to complete ... All CUs completed! Closed session, exiting now ...
5.1.2. Logging and Debugging¶
Since working with distributed systems is inherently complex and much of the
complexity is hidden within RADICAL-Pilot, it is necessary to do a lot of
internal logging. By default, logging output is disabled, but if something goes
wrong or if you’re just curious, you can enable the logging output by setting
the environment variable
RADICAL_PILOT_VERBOSE to a value between CRITICAL
(print only critical messages) and DEBUG (print all messages). For more details
on logging, see under ‘Debugging’ in chapter Developer Documentation.
Give it a try with the above example:
RADICAL_PILOT_VERBOSE=DEBUG python simple_bot.py