Store large data or a service connection per Flask session

Question

I'm writing a small Flask application and am having it connect to Rserve using pyRserve. I want every session to initiate and then maintain its own Rserve connection.

Something like this:

session['my_connection'] = pyRserve.connect()

doesn't work because the connection object is not JSON serializable. On the other hand, something like this:

flask.g.my_connection = pyRserve.connect()

doesn't work because it does not persist between requests. To add to the difficulty, it doesn't seem as though pyRserve provides any identifier for a connection, so I can't store a connection ID in the session and use that to retrieve the right connection before each request.

Is there a way to accomplish having a unique connection per session?

Kancler

Why do you need to use the same connection for a session? – Helmholtz

Because I need objects in the R namespace to persist for the same user during a session (but not be visible / accessible to other users). For instance, a user may load some data and fit a model - I want to be able to access that model (without refitting it) on other pages (ie. after other Flask requests have been made). – Kancler

I see. I'm not certain I truly need a re-usable connection per user. My only requirement is that a user's R connection/session be able to access R objects created using previous requests by that user. I suppose a workable solution might be to have an R connection save the current R workspace to the server, save the ID of that workspace as a cookie, and upon a new request, have a new R connections read that workspace back... – Kancler

Take a look at DeployR (deployr.revolutionanalytics.com) - it adds APIs and additional functionality on top of Rserve that makes it easy to manage this type of requirement. – Hallock

@Hallock I considered this - although looks like there are only client libraries for Java, Javascript, and .NET. I'm restricted to python... – Kancler

You can call the API directly, without using the client libraries. See deployr.revolutionanalytics.com/documents//dev/api-doc/guide/… or ask a question at groups.google.com/forum/#!forum/deployr – Hallock

Answer 1

The following applies to any global Python data that you don't want to recreate for each request, not just rserve, and not just data that is unique to each user.

We need some common location to create an rserve connection for each user. The simplest way to do this is to run a multiprocessing.Manager as a separate process.

import atexit
from multiprocessing import Lock
from multiprocessing.managers import BaseManager
import pyRserve

connections = {}
lock = Lock()


def get_connection(user_id):
    with lock:
        if user_id not in connections:
            connections[user_id] = pyRserve.connect()

        return connections[user_id]


@atexit.register
def close_connections():
    for connection in connections.values():
        connection.close()


manager = BaseManager(('', 37844), b'password')
manager.register('get_connection', get_connection)
server = manager.get_server()
server.serve_forever()

Run it before starting your application, so that the manager will be available:

python rserve_manager.py

---

We can access this manager from the app during requests using a simple function. This assumes you've got a value for "user_id" in the session (which is what Flask-Login would do, for example). This ends up making the rserve connection unique per user, not per session.

from multiprocessing.managers import BaseManager
from flask import g, session

def get_rserve():
    if not hasattr(g, 'rserve'):
        manager = BaseManager(('', 37844), b'password')
        manager.register('get_connection')
        manager.connect()
        g.rserve = manager.get_connection(session['user_id'])

    return g.rserve

Access it inside a view:

result = get_rserve().eval('3 + 5')

---

This should get you started, although there's plenty that can be improved, such as not hard-coding the address and password, and not throwing away the connections to the manager. This was written with Python 3, but should work with Python 2.

Inquisitive