Take DB downtime into account

[bkonkle]

When a database in the pool is down, temporarily remove it from the pool but periodically re-check the connection. If it comes back up, add the database back to the pool.

[bkonkle]

It appears that psycopg2.OperationalError (the exception raised when the database cannot be connected to) is not caught by the try...except block inside django.core.handlers.base.BaseHandler.get_response, so I'm not able to use django.core.signals.got_request_exception to take that database out of the pool. I'm trying to find another way, but it looks like I may be going down the wrong path. I'm open to suggestions on other ways to handle this.

[bkonkle]

I could probably create a new database backend that subclasses DatabaseWrapper and wraps ._cursor() in a try...except block. I'd have to make a new backend for every existing backend, though, and that would be messy. I want something more elegant. Back to the drawing board...

[wolph]

Wrapping the cursor (or even ... you could monkeypatch it) is probably the easiest way.

I have looked at this issue a couple of months ago when writing django-statsd (http://pypi.python.org/pypi/django-statsd/) and it seemed that monkeypatching is the best way to go here.

But if you know any good way to do it, please let me know.