Skip to main content

A minimal cluster computing framework

Project description

minimal” here means minimal dependency or platform requirements, as well as its nature of “minimum viable product”. It’s mainly for tackling straightforward “embarrassingly parallel” tasks using multiple commodity machines, also a good choice for experimental and learning purposes. The idea came from Eli Bendersky’s blog .

minimalcluster is built only using plain Python and its standard libraries (mainly multiprocessing). This brings a few advantages, including

  • no additional installation or configuration is needed

  • 100% cross-platform (you can even have Linux, MacOS, and Windows nodes within a single cluster).

This package can be used with Python 2.7+ or 3.6+. But within a cluster, you can only choose a single version of Python, either 2 or 3.

For more frameworks for parallel or cluster computing, you may also want to refer to Parallel Processing and Multiprocessing in Python .



Usage & Examples

Step 1 - Install this package

pip install minimalcluster

Step 2 - Start master node

Open your Python terminal on your machine which will be used as Master Node, and run

from minimalcluster import MasterNode

your_host = '<your master node hostname or IP>'
# or use '' if you have high enough privilege
your_port= <port to use>
your_authkey = '<the authkey for your cluster>'

master = MasterNode(HOST = your_host, PORT = your_port, AUTHKEY = your_authkey)

Please note the master node will join the cluster as worker node as well by default. If you prefer otherwise, you can have argument if_join_as_worker in start_master_server() to be False. In addition, you can also remove it from the cluster by invoking master.stop_as_worker() and join as worker node again by invoking master.join_as_worker().

Step 3 - Start worker nodes

On all your Worker Nodes, run the command below in your Python terminal

from minimalcluster import WorkerNode

your_host = '<your master node hostname or IP>'
your_port= <port to use>
your_authkey = '<the authkey for your cluster>'
N_processors_to_use = <how many processors on your worker node do you want to use>

worker = WorkerNode(your_host, your_port, your_authkey, nprocs = N_processors_to_use)


Note: if your nprocs is bigger than the number of processors on your machine, it will be changed to be the number of processors.

After the operations on the worker nodes, you can go back to your Master node and check the list of connected Worker nodes.


Step 4 - Prepare environment to share with worker nodes

We need to specify the task function (as well as its potential dependencies) and the arguments to share with worker nodes, including

Environment: The environment is simply the codes that’s going to run on worker nodes. There are two ways to set up environment. The first one is to prepare a separate .py file as environment file and declare all the functions you need inside, then use master.load_envir(‘<path of the environment file>’) to load the environment. Another way is for simple cases. You can use master.load_envir(‘<your statements>’, from_file = False) to load the environment, for example master.load_envir(“f = lambda x: x * 2”, from_file = False).

Task Function: We need to register the task function using master.register_target_function(‘<function name>’), like master.register_target_function(“f”). Please note the task function itself must be declared in the environment file or statement.

Arguments: The argument must be a list. It will be passed to the task function. Usage: master.load_args(args). Note the elements in list args must be unique.

Step 5 - Submit jobs

Now your cluster is ready. you can try the examples below in your Python terminal on your Master node.

Example 1 - Estimate value of Pi

envir_statement = '''
from random import random
example_pi_estimate_throw = lambda x: 1 if (random() * 2 - 1)**2 + (random() * 2 - 1)**2 < 1 else 0
master.load_envir(envir_statement, from_file = False)

N = int(1e6)

result = master.execute()

print("Pi is roughly %f" % (4.0 * sum([x2 for x1, x2 in result.items()]) / N))

Example 2 - Factorization

envir_statement = '''
# A naive factorization method. Take integer 'n', return list of factors.
# Ref:
def example_factorize_naive(n):
    if n < 2:
        return []
    factors = []
    p = 2
    while True:
        if n == 1:
            return factors
        r = n % p
        if r == 0:
            n = n / p
        elif p * p >= n:
            return factors
        elif p > 2:
            p += 2
            p += 1
    assert False, "unreachable"

#Create N large numbers to factorize.
def make_nums(N):
    nums = [999999999999]
    for i in range(N):
        nums.append(nums[-1] + 2)
    return nums

master.load_envir(envir_statement, from_file = False)

result = master.execute()

for x in result.items()[:10]: # if running on Python 3, use `list(result.items())` rather than `result.items()`

Example 3 - Feed multiple arguments to target function

It’s possible that you need to feed multiple arguments to target function. A small trick will be needed here: you need to wrap your arguments into a tuple, then pass the tuple to the target function as a “single” argument. Within your argument function, you can “unzip” this tuple and obtain your arguments.

envir_statement = '''
f = lambda x:x[0]+x[1]
master.load_envir(envir_statement, from_file = False)

master.load_args([(1,2), (3,4), (5, 6), (7, 8)])

result = master.execute()


Step 6 - Shutdown the cluster

You can shutdown the cluster by running


Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

minimalcluster-0.1.0.dev13-py2.py3-none-any.whl (14.6 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page