Newer versions of Celery cache the broker URL. This was preventing any

new URL changes from taking effect until the web server was restarted.
We now lazily-construct the Celery instance when we request it, and we

invalidate the instance if the broker URL has changed.

Switched back-and-forth between valid and invalid broker URLs. Saw that

the worker status only returned successful results when using the valid

broker URL, rather than fetching from whatever URL was used in the first

request.