Skip to content


Subversion checkout URL

You can clone with
Download ZIP
An async programming framework with a blocking look-alike syntax.
Pull request Compare This branch is 4 commits ahead, 112 commits behind saucelabs:master.
Fetching latest commit...
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

monocle - An async programming framework with a blocking look-alike syntax.

By Greg Hazel and Steven Hazel.

monocle straightens out event-driven code using Python's generators. It aims to be portable between event-driven I/O frameworks, and currently supports Twisted and Tornado.

It's for Python 2.5 and up; the syntax it uses isn't supported in older versions of Python. (Versions before 2.7 require the ordereddict module.)

A Simple Example

Here's a simple monocle program that runs two concurrent lightweight processes (called "o-routines") using Tornado's event loop. One is an HTTP server, and the other makes an HTTP request:

import monocle

from monocle import Return
from monocle.stack import eventloop
from import add_service
from import HttpClient, HttpHeaders, HttpServer

def hello_http(req):
    content = "Hello, World!"
    headers = HttpHeaders()
    headers['Content-Length'] = len(content)
    headers['Content-Type'] = 'text/plain'
    yield Return(200, headers, content)

def request():
    client = HttpClient()
    resp = yield client.request('')
    print resp.code, resp.body

add_service(HttpServer(hello_http, 8088))


It's important that code be dapper and well-dressed, so if you prefer, you can don the monocle and use this handy shortcut for @monocle.o:

from monocle import _o

def request():
    client = HttpClient()
    resp = yield client.request('')
    print resp.code, resp.body

It's true, this violates Python's convention that underscores indicate variables for internal use. But rules are for breaking. Live a little.

The Big Idea

Event-driven code can be efficient and easy to reason about, but it often splits up procedures in an unpleasant way. Here's an example of a blocking function to read a request from a user, query a database, and return a result:

def do_cmd(conn):
    cmd = conn.read_until("\n")
    if cmd.type == "get-address":
        user = db.query(cmd.username)
        conn.write("unknown command")

Here's the same thing in event-driven style, using callbacks:

def get_cmd(conn):
    conn.read_until("\n", callback=handle_cmd)

def handle_cmd(conn, cmd):
    if cmd.type == "get-address":
        # keep track of the conn so we can write the response back!
        def callback(result):
            handle_user_query_result(conn, result)
        db.query(cmd.username, callback)
        conn.write("unknown command")

def handle_user_query_result(conn, user):

What started out as a single function in the blocking code has expanded here into four functions (counting the callback closure that captures conn in handle_cmd). In real event-driven code, this kind of thing happens a lot. Any time we want to do I/O, we have to register a new handler and return back out to the event loop to let other things happen while we wait for the I/O to finish. It would be nice if we had some way to tell the event loop to call back into the middle of our function, so we could just continue where we left off.

Fortunately, Python has a mechanism that lets us do exactly that, called generators. Monocle uses generators to straighten out event-driven code.

Here's the monocle equivalent of the event-based code above:

def do_cmd(conn):
    cmd = yield conn.read_until("\n")
    if cmd.type == "get-address":
        user = yield db.query(cmd.username)
        yield conn.write(user.address)
        yield conn.write("unknown command")

It's event-driven for efficient concurrency, but otherwise looks a lot like the original blocking code. Each time you see the word yield in the code above, the o-routine is returning back up to the event loop and waiting to be called back when the I/O it requested completes.

This approach is a kind of cooperative concurrency that makes for simpler code than callback-based event-driven code, but which we think is easier to reason about than multi-threaded code.

A word about the word yield

In ordinary Python generators, the norm is to think of yield as in crops: the generator yields a value. In monocle o-routines, it's helpful to think of yield as in traffic. yield in an o-routine means "yield to other o-routines until we finish reading 10 bytes".

Related Work

monocle is similar to, and takes inspiration from:

  • Twisted's inlineCallbacks
  • BitTorrent's yielddefer (used in the 5.x mainline client)
  • diesel
  • Go's goroutines (and CSP generally)
  • Haskell's I/O monad
  • eventlet
Something went wrong with that request. Please try again.