Pure Python & Flask server-side event source

index.html

<script>
var stream = new EventSource('/api/stream');
stream.onmessage = function(e) {
    console.info(e.data);
};
</script>

server.py

from queue import Queue
from flask import Flask, render_template, request

app = Flask(__name__)
app.debug = True

queue = Queue()

def event_stream():
    while True:
    	# Question: how to break out of this loop when the app is killed?
        message = queue.get(True)
        print("Sending {}".format(message))
        yield "data: {}\n\n".format(message)


@app.route('/')
def hello():
    return render_template('index.html')


@app.route('/api/stream')
def stream():
    return flask.Response(event_stream(), mimetype="text/event-stream")


@app.route('/api/post', methods=['GET'])
def api_parse_sentence():
    queue.put(request.args.get('sentence'))
    return "OK"


if __name__ == '__main__':
    app.run(threaded=True)

queue.get accepts a second parameter as timeout. Maybe a 500ms timeout would be enough to notice the restart efficiently enough?

Was amusing myself today looking around for Python server side event examples and came across this.

Currently, only one of the /api/stream will get the message since you are sharing the same queue in event_stream().

The below makes a queue for each subscription and broadcasts to all of them (granted, only safe for low request rates):

from queue import Queue
from flask import Flask, render_template, request, Response

app = Flask(__name__)
app.debug = True

QUEUES = []


def event_stream(q):
    while True:
        # Question: how to break out of this loop when the app is killed?
        message = q.get(True)
        print("Sending {}".format(message))
        yield "data: {}\n\n".format(message)


@app.route('/')
def hello():
    return render_template('index.html')


@app.route('/api/stream')
def stream():
    q = Queue()
    QUEUES.append(q)

    return Response(event_stream(q), mimetype="text/event-stream")


@app.route('/api/post', methods=['GET'])
def api_parse_sentence():
    for q in QUEUES:
        q.put(request.args.get('sentence'))

    return "OK"


if __name__ == '__main__':
    app.run(host='0.0.0.0', port=3000, debug=True, threaded=True)

@eliwjones That's a very fair point. Hadn't encountered that yet, but makes total sense. Thanks for the revision.

You're also right that this won't work for more than a very simple test setup (in case of your example, QUEUES would just keep on growing and all queues that won't be listened to anymore will keep getting saturated). So let that be a warning to anyone wanting to use this snippet :)