Real Time Notifications with Pyramid and ZeroMQ

Sooner or later I had to implement real time notifications on my latest project Floresta. The problem was that most of the code is synchronous. I think the ideal solution will be to run aiohttp on a separate instace or in a separate thread, but since I’m still on python 2.7 that was not possible. After researching on the subject I opted to enable gevent on my current uWSGI setup and add ZeroMQ to the equation.

So the first thing I did was install pyzmq, gevent and ZeroMQ. On mac is easy just use homebrew and pip. Make sure you install ZeroMQ version 4.1.5 or higher and pymzq 15.4.0 or higher.

On Debian I had to do the following:

Install libsodium first

git clone git://
cd libsodium
./configure && make check 
sudo make install 
sudo ldconfig

Then compile ZeroMQ

git clone
cd zeromq-4.1.X
./configure && make check 
sudo make install 
sudo ldconfig

On the uWSGI side I had to add these lines to my .ini file

gevent = 100
gevent-monkey-patch = true
enable-threads = true

Now the view

# coding=utf-8
from pyramid.response import Response
from pyramid.view import view_config, notfound_view_config,forbidden_view_config
from pyramid.httpexceptions import (
from sqlalchemy.orm.exc import NoResultFound,MultipleResultsFound
import logging
import os
import datetime
import time
import json 
import threading
import  zmq
log = logging.getLogger(__name__)
sock = "ipc:///tmp/zmq.test"
context = zmq.Context()
pub_socket = context.socket(zmq.PUB)
pub_lock = threading.Lock()
def message_generator():
        socket2 = context.socket(zmq.SUB)
        #make sure to connect and not bind. You can only have one connect and multiple binds
        you can filter messages based on user id from a DB table as in
        socket2.setsockopt(zmq.SUBSCRIBE, userid)
        socket2.setsockopt(zmq.SUBSCRIBE, ')  
    except zmq.error.ZMQError:
        console.log("socket already in use, try restarting it")
        # run forever
        while True:
                msg = socket2.recv(zmq.NOBLOCK)
                # break out of the loop
                if msg == "EXIT":
                    console.log("exiting process.")
                yield "data: %s\n\n" % json.dumps({"message": msg})
                console.log("sending message")
                console.log("nothing send..")
    except GeneratorExit:
#the url that streams the events to the browser
def _orders_events(request):
    headers = [("Content-Type", "text/event-stream"),
                       ("Cache-Control", "no-cache")]
    response = Response(headerlist=headers)
    response.app_iter = message_generator(userid)
    return response
#the url that publishes messages to the subscriber.
def push(request):
    msg = json.loads(request.body)["message"]
    with pub_lock:
    return Response()

The message_generator def will run forever inside your app. The way to stop it is to send a message with the string “Exit”.

Once you have this view in place, all you have to do is connect to the events url from javascript with the following code.

var source = null;
$(function() {
    source = new EventSource("/events");
    source.addEventListener("message", messageReceived, false);
    source.onerror = eventSourceErrorFunction;
    var eventSourceErrorFunction = function(event){
        if (event.eventPhase == EventSource.CLOSED) {
            console.log("Event Source Closed");
function messageReceived(event) {
console.log("message arrived: " + msg)

Send test messages from the Javascript console

        url: "/events/push",
        data: JSON.stringify({message: "Remote message"}),
        type: "post",
        success: function() {
            console.log("message sent!");

If you plan on placing Nginx in front of uWSGI don’t forget to turn off uwsgi_buffering and set the uwsgi_read_timeout to 300 otherwise the messages will get stuck.