Is it correct to modify `django.db.connections.databases` dynamically to multiple databases? Or is it better to deploy a separate API per client?

This is my first time developing a multi-tenant SaaS application in Django, in this SaaS each company has its own PostgreSQL database, and these databases are created dynamically when a company registers. I cannot predefine all databases in settings.DATABASES, as companies can register at any time without requiring a server restart.

My current solution uses Middleware to detect the company from the subdomain or a JWT token and then modify connections.databases at runtime to configure the connection to the company's database:

import redis
from django.db import connections
from django.core.exceptions import ImproperlyConfigured
from django.utils.connection import ConnectionDoesNotExist
from rest_framework_simplejwt.authentication import JWTAuthentication
from myapp.models import Company  # Company model stored in the global database

class CompanyDBMiddleware:
    def __init__(self, get_response):
        self.get_response = get_response
        self.jwt_authenticator = JWTAuthentication()
        self.cache = redis.Redis(host='localhost', port=6379, db=0)

    def __call__(self, request):
        company_db = self.get_database_for_company(request)

        if not company_db:
            raise ImproperlyConfigured("Could not determine the company's database.")

        # Register connection only if it does not exist in `connections.databases`
        if company_db not in connections.databases:
            connections.databases[company_db] = {
                'ENGINE': 'django.db.backends.postgresql',
                'NAME': company_db,
                'USER': 'postgres',
                'PASSWORD': 'your_password',
                'HOST': 'localhost',
                'PORT': '5432',
                'CONN_MAX_AGE': 60,  # To avoid opening and closing connections on each request
            }

        request.company_db = company_db
        response = self.get_response(request)

        # Close connection after the response
        try:
            connections[company_db].close()
        except ConnectionDoesNotExist:
            pass

        return response

    def get_database_for_company(self, request):
        subdomain = request.get_host().split('.')[0]
        company_db = None

        cache_key = f"company_db_{subdomain}"
        company_db = self.cache.get(cache_key)

        if company_db:
            return company_db.decode("utf-8")

        try:
            company = Company.objects.using('default').get(subdomain=subdomain, active=True)
            company_db = company.db_name
            self.cache.setex(cache_key, 300, company_db)  # Cache the database name for 5 minutes
            return company_db
        except Company.DoesNotExist:
            return None

My questions are:

✅ Is it correct to modify connections.databases dynamically on each request to handle multiple databases?
✅ Is there a better way to do this in Django without restarting the application when registering new databases?
✅ How does this practice affect performance in environments with load balancing and multiple Django instances?
✅ Would it be better to deploy a separate API per client in its own Django container?
✅ I am considering giving each client their own portal on a separate domain and only deploying their frontend in a container while keeping a centralized API. Is this approach more efficient?

I appreciate any recommendations on best practices or potential issues with this approach. Thanks! 😊

Вернуться на верх