Cached Session may cause losing a session, when it failed to connect to cache backend

[Kenial Sookyum Lee]

Some cache backends (AFAIK Memcached and Redis) has a feature to ignore connection timeout exception, in order to ensure Django application working even if cache has failed. This can lead a subtle bug, deleting a session cookie. Following steps are reproduce this bug:

• Set cached session for Django session (refer to ​https://docs.djangoproject.com/en/2.0/topics/http/sessions/#using-cached-sessions)
• On Django app, try to log in or set a language in order to make a session cookie. At this step, cache nodes should be available.
• To set cache nodes unavailable, by changing to wrong node name or turning the nodes off.
• Reload a page on Django app. Django app responds back with this HTTP header, which deletes a session cookie:

  Set-Cookie:  sessionid=""; expires=Thu, 01-Jan-1970 00:00:00 GMT; Max-Age=0; Path=/
  

The problematic codes are at django/contrib/sessions/middleware.py:37 :

 def process_response(self, request, response):
        """
        If request.session was modified, or if the configuration is to save the
        session every time, save the changes and set a session cookie or delete
        the session cookie if the session has been emptied.
        """
        try:
            accessed = request.session.accessed
            modified = request.session.modified
            empty = request.session.is_empty()
        except AttributeError:
            pass
        else:
            # First check if we need to delete this cookie.
            # The session should be deleted only if the session is entirely empty
            if settings.SESSION_COOKIE_NAME in request.COOKIES and empty:     # NEED TO UPDATE!
                response.delete_cookie(
                    settings.SESSION_COOKIE_NAME,
                    path=settings.SESSION_COOKIE_PATH,
                    domain=settings.SESSION_COOKIE_DOMAIN,
                )
...

I guess the initial intention was to delete a session if there is no values in it. However, it happens that code execution reaches at :37 that code without exception, even if cache nodes are unavailable. The reason is, I already explained above, they just work that way even if the cache backend failed.

I first met this bug while I was testing failover of AWS Elasticache (Redis). I was in testing of failover scenario, but Django application got me logged out repeatedly, even though session data itself is remaining in the cache replica. (it should, because it was doing failover, not reboot)

IMHO, before checking empty of session data, there should be a logic to check cache backend is actually available. I found out request.session.cache_key can do that function, but it looks less explicit. Please show be a better way to do this, if you have one.

fyi. Configuration is: Django 1.11, MySQL 5.6.35, mysqlclient 1.3.12, Redis 3.2.7 (x64), and django-redis 4.9.0. I found out this bug on Django 1.11, but it has been changed since. So I believe this bug is applied to Django 2.x as well.

[Kenial Sookyum Lee]

workaround patch

[Kenial Sookyum Lee]

When I tested the settings with SESSION_ENGINE = "django.contrib.sessions.backends.cached_db" (ie. 'write-through cache'), I didn't get through this bug. Configuration is: Django 1.11, MySQL 5.6.35, mysqlclient 1.3.12, Redis 3.2.7 (x64), django-redis 4.9.0.