errand-boy 0.3.5

Establish a connection to the errand-boy deamon to execute commands without the memory overhead incurred by os.fork().

What does it do?

Uses Python multiprocessing to maintain a pool of worker processes used to execute arbitrary terminal commands.

Why not use subprocess.Popen()?

Under the hood subprocess.Popen() uses os.fork(), which copies the currently running process’ memory before launching the command you want to run. If your process uses a lot of memory, such as a Celery worker using an eventlet pool, this can cause a “Cannot allocate memory” error.

errand-boy still uses subprocess.Popen(), but tries to keep a low memory footprint. Your celery greenthread workers can communicate with it via asynchronous network calls.

Further reading:

1. http://stackoverflow.com/a/13329386/241955

2. http://stackoverflow.com/a/14942111/241955

Setup

Install:

pip install errand-boy

# optional pip install setproctitle

Usage

Run tests:

cd errand-boy/
python -m unittest discover

Run server:

python -m errand_boy.run

Run client (useful for testing/debugging):

python -m errand_boy.run 'ls -al'

Use the client in your code:

from errand_boy.transports.unixsocket import UNIXSocketTransport


errand_boy_transport = UNIXSocketTransport()

stdout, stderr, returncode = errand_boy_transport.run_cmd('ls -al')

print stdout
print stderr
print returncode

Use a subprocess.Popen-like interface:

from errand_boy.transports.unixsocket import UNIXSocketTransport


errand_boy_transport = UNIXSocketTransport()

# Attribute accesses and function calls on objects retrieved via a session
# result in a call to the errand-boy server, unless that object is a string
# or number type.
with errand_boy_transport.get_session() as session:
    subprocess = session.subprocess

    # Here, subprocess.Popen is actually a reference to the actual objects
    # on the errand-boy server. subprocess.PIPE is an int.
    process = subprocess.Popen('ls -al', stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True, close_fds=True)

    # Here, process_stdout and process_stderr are strings and returncode is
    # an int, so their actual values are returned instead of references to
    # the remote objects. This means it's safe to use these values later on
    # outside of the current session.
    process_stdout, process_stderr = process.communicate()
    returncode = process.returncode

print stdout
print stderr
print returncode

# Since the session has been closed, trying this will result in an error:
print process.returncode
# raised errand_boy.exceptions.SessionClosedError()

Run load tests:

python -m errand_boy.run --max-accepts=0

pip install Fabric locustio
cd errand-boy/
fab locust_local

Does it work in other languages?

The client/server use an HTTP-inspired protocol, but the data that’s sent back and forth is currently serialized using Python’s Pickle format. Support could be added for other serialization types though.

Development

Further reading:

• http://stackoverflow.com/questions/18414020/memory-usage-keep-growing-with-pythons-multiprocessing-pool