pretzel 1.0.10

Pretzel

Is an asynchronous application framework for python

Features

• C# like async/await(async/yield) paradigm for asynchronous programming (monad base)

• Cool asynchronous I/O loop implementation

• Uniform asynchronous stream implementation for sockets and pipes

• Interact with subprocesses asynchronously

• Greenlet support (but not required)

• Remote code executing over ssh or in child process (with only requirements python and ssh)

• Python 2/3, PyPy (starting from 2.0) compatible

• Asynchronous python shell python -mpretzel.apps.shell (requires greenlet)

Installation

As git submodule: git submodule add git://github.com/aslpavel/pretzel.git <path_to_submodule> Pip from git: pip install git+git://github.com/aslpavel/pretzel-pkg.git Pip from PyPI pip install pretzel

Approach

Usage of asynchronous functions is similar to C# async/await but instead of async attribute it uses @async decorator and instead of await keyword, yield is used. Internally unit of asynchrony is implemented as continuation monad Cont with embedded Result monad (similar to Haskell’s Cont and Either monads) as its value. One important difference of Cont monad from C# Task object, is that Task represents already running asynchronous operation, but continuation monad is a sequence of computations, and this computations are not started. .future() method on instance of Cont can be used to create Task like object. To use this library you don’t have to understand notion of the monad. Simple asynchronous function would look like this.

from pretzel.monad import async
from pretzel.core import sleep

@async
def print_after(delay, *args, **kwargs):
  """Calls print function after the lapse of `delay` sedonds.
  """
  yield sleep(delay)  # execution will be resumed in delay seconds
  print(*args, **kwargs)

To return something meaningful in python3 you can just use return keyword, but in python2 you have to use do_return function (it will also work in python3) as return with value cannot be used inside a generator function. Result of such asynchronous function is again a continuation monad, if exception is thrown during execution of its body, it is marshaled to receiver of the result and can be processed correctly. For example.

@async
def process_error():
  @async
  def throw_after(delay, error):
    yield sleep(delay)
    raise error

  try:
    yield throw_after(1, ValueError('test error'))
  except ValueError as error:
    # process error in a meaningfull way
  do_return('done')  # exectly equivalent to: return 'done'

Asynchronous values (continuation monads) can be composed with two helper functions async_all and async_any.

@async
def composition_example():
  yield async_all([sleep(1), sleep(2)])  # will be resumed in 2 seconds
  yield async_any([sleep(1), sleep(2)])  # will be resumed in 1 sedond

  result_all = yield async_all([func1(), func2()])  # = (result1, result2)
  reuslt_any = yield async_any([func1(), func2()])  # = result1 | result2

Cont monad can also be called with callback function as its argument, in this case, on completion of asynchronous operation, callback will be called with Result monad. If callback function is not specified default, then default continuation callback will be used which only reports errors if any.

>>> sleep(1)(print)
Result(val:1374307530.015137)
>>> sleep(None)()
[continuation] error in coroutine started from
  File "<console>", line 1, in <module>
`-------------------------------------------------------------------------------
Host   : fiend
Process: 13492
Error  : TypeError("unsupported operand type(s) for +: 'float' and 'NoneType'")

Traceback (most recent call last):
  File "./pretzel/monad/do.py", line 26, in do_block
    return value(block(*args, **kwargs))
  File "./pretzel/core/core.py", line 118, in sleep
    do_done(self.time_queue.on(time() + delay))
TypeError: unsupported operand type(s) for +: 'float' and 'NoneType'

Inside body of asynchronous function you can yield not only Cont monad directly, but any object implementing .__monad__() method which returns Cont monad. There are many such types in this library, for example Event

@async
def func():
  print(1)
  yield event
  print(2)
  print((yield event))
event = Event()
func()()     # 1 is printed
event('e0')  # 2 is printed
event('e1')  # 'e1' is printed

Main loop

Core class implements I/O loop, and it is used internally to implement asynchronous streams, timers and more. Previously used sleep function will work correctly only in presence of running I/O loop. Simplest way to intialize and use Core object is to use @app decorator.

"""Minimal pretzel application

Sleeps for one second, then prints 'done' and exits.
"""
from pretzel.app import app
from pretzel.core import sleep

@app
def main():
  yield sleep(1)
  print('done')

if __name__ == '__main__':
  main()

See Core’s init module for full list of available functions and their descriptions.

Remoting

Main reason for creation of this framework was to execute code on a set of machines via ssh connection. It is achieved by usage of SSHConnection class. SSHConnection is a callable object which returns proxy object for its argument. You can call proxy object, get its attributes or items proxy[item], result of such operations is again a proxy object with this embedded operations. Proxy implements monad interface, and to get result of embedded chain of operations you can yield it inside asynchronous function. In this example we create proxy for os.getpid function, call it and then execute on remote process by yielding it. There is no need for pretzel to be installed on remote machine.

import os
from pretzel.app import app
from pretzel.remoting import SSHConnection

@app
def main():
  """Connect to localhost via ssh and print remote process's pid

  Note:
    You have to be able to login to the remote host without
    entering any password (by means of ssh keys) otherwise
    connecition will fail.
  """
  with (yield SSHConnection('localhost')) as conn:
    print((yield conn(os.getpid)()))

if __name__ == '__main__':
  main()

Connection can marshal any pickle-able object, or Sender object plus any object which is reducible to set of pickle-able and Sender objects. Proxy and Connection itself are examples of such objects. You can also create proxy object from any arbitrary object with proxify or proxify_func.

import os
from pretzel.app import app
from pretzel.remoting import SSHConnection, proxify

class Remote(object):
  """Object which will be used remotely
  """
  def __init__(self):
    self.value = 0

  def next(self):
    self.value += 1
    return self.value

  def getpid(self):
    return os.getpid()

@app
def main():
  with (yield SSHConnection('localhost')) as conn:
    with (yield proxify(conn(Remote)())) as o:  # remote object proxy
      print(os.getpid(), (yield o.getpid()))    # prints two different pids
      print((yield o.next()))  # prints 1
      print((yield o.next()))  # prints 2

if __name__ == '__main__':
  main()

But Cont monad is not marshallable, that is why there is special operation on proxy object ~ which is equivalent to yield inside asynchronous function. Here is an example of remote execution of asynchronous function.

from pretzel.app import app
from pretzel.process import process_call
from pretzel.remoting import SSHConnection

@app
def main():
  """Execute 'ls' on remote machine and show result of the execution
  """
  with (yield SSHConnection('localhost')) as conn:
    out, err, code = yield ~conn(process_call)('ls')
    print(out.decode())

if __name__ == '__main__':
  main()

There is also a way to work with multiple connections as if it one, by means of composite_ssh_conn. It accepts list of hosts and returns composite connection, which behaves as ordinary connection but returns set of results.

import os
from pretzel.app import app
from pretzel.remoting import composite_ssh_conn

@app
def main():
  hosts = ['localhost', 'localhost']
  with (yield composite_ssh_conn(hosts)) as conns:
    result = yield conns(os.getpid)()
    print(result)  # List(25163, 25162) - iterable object of pids

if __name__ == '__main__':
  main()

Remoting submodule can be used as workaround for python’s GIL, in a similar fashion to multiprocessing module. You can use ForkConnection (or composite_fork_conn) which behaves as SSHConnection but instead of connecting via ssh, it just spawns new process.

import time
from pretzel.app import app
from pretzel.remoting import composite_fork_conn

def computation_heavy_task():
  """Some computation intensive task
  """
  start_time = time.time()
  time.sleep(10)
  stop_time = time.time()
  return int(stop_time - start_time)

@app
def main():
  with (yield composite_fork_conn(10)) as conns:  # create 10 connections
    result = yield conns(computation_heavy_task)()
    print(result)  # prints List(10, 10, 10, 10, 10, 10, 10, 10, 10, 10)

if __name__ == '__main__':
  main()

Examples

• Simple echo server

• Cat remote file over ssh