Django - Losing connection with database on multi-tenant app

On my Django app I use a multi-tenant approach with isolated databases. It works well but because it relies on subdomains for each tenant, which is not scalable, I’m trying to change that. To achieve this functionality I’m trying to use middlewares and sessions to retrieve the tenant from the username and use it to set a local variable for the database routers. The logic is this:

If the user is not logged, the BeforeLoginMiddleware activates and retrieves the tenant name from the user. So username@tenant1 will set tenant1 to the session. Here’s the code:

import threading
from users.forms import LoginForm

Thread_Local = threading.local()

class BeforeLoginMiddleware:

    def __init__(self, get_response):
        self.get_response = get_response

    def __call__(self, request):

        if request.path == '/login/':

            form = LoginForm(request.POST)

            if form.is_valid():

                complete_username = form.cleaned_data.get('username')
                current_db = complete_username.split('@')[1]
                request.session['current_db'] = current_db
                request.session.modified = True

        response = self.get_response(request)

        return response

If the user is already logged, a second middleware will retrieve the tenant data from the session and use it to define the Thread_Local variable that is called on a function used on the database routers:

class AppMiddleware:

    def __init__(self, get_response):
        self.get_response = get_response

    def __call__(self, request):

        current_db = request.session.get('current_db')
        setattr(Thread_Local, 'current_db', current_db)
        response = self.get_response(request)

        return response

def get_current_db_name():
    return getattr(Thread_Local, 'current_db', None)

Here’s the routers.py file:

class AppRouter:

    def db_for_read(self, model, **hints):
        return get_current_db_name()

    def db_for_write(self, model, **hints):
        return get_current_db_name()

    def allow_relation(self, *args, **kwargs):
        return True

    def allow_syncdb(self, *args, **kwargs):
        return None

    def allow_migrate(self, *args, **kwargs):
        return None

This is the middleware setting on my app:

MIDDLEWARE = [
    'debug_toolbar.middleware.DebugToolbarMiddleware',
    'django.middleware.security.SecurityMiddleware',
    'django.contrib.sessions.middleware.SessionMiddleware',
    'django.middleware.common.CommonMiddleware',
    'django.middleware.csrf.CsrfViewMiddleware',
    'myapp.middleware.BeforeLoginMiddleware',
    'django.contrib.auth.middleware.AuthenticationMiddleware',
    'django.contrib.messages.middleware.MessageMiddleware',
    'django.middleware.clickjacking.XFrameOptionsMiddleware',
    'myapp.middleware.AppMiddleware',
]

This works as expected but at random (or at least it appear to be at random) it loses connection with the database and I assume that, since it can’t retrieve the session data, redirects the user to the login page again (I’m also using the login_required decorator). Sometimes I can navigate to several different pages and everything is fine and then it disconnects. Other times it disconnects when navigating to the first page after the homepage, after login. This also occurs if the page is idle for about 1 minute, which makes no sense because my SESSION_COOKIE_AGE setting is 1200.

Thing is, I have no idea what is causing this because there are no errors. The only thing I’ve noticed that is consistent when this happens, it’s a 302 status registered in the network tab on the browser (which is the redirect to the login page) and the message Broken pipe from (‘127.0.0.1’, 61980) in the terminal with this random code at the end.

What I’ve tried so far, without any change in the behavior described above:

  • A LOT of different settings for the middleware. For example: at first I wrote a single middleware with all the logic and put it in the end of the middleware list. I thought that this could be causing the issue so I separated the logic. Another try was to use request.user.is_authenticated in the middleware instead of using the session data.
  • Based on various posts on various forums, the issue could be reIated to the browser sending a request but not waiting for the response and closing the connection. Because this could be caused by JS scripts, i disabled every JS script on the app (there are just a few). Not really an option but I think it excludes the possibility of being the cause.
  • Clear the browser cache / cookies entirely.
  • Change the SESSION_SAVE_EVERY_REQUEST setting to True.
  • Change the SESSION_COOKIE_AGE setting to 120000 (I think it makes no sense to increase this number, but at this point, I’m willing to try anything).

I’m honestly running out of options, so I appreciate any help or advice on how to pinpoint the problem or alternative approaches to achieve this. Thank you!