Ceph » Orchestrator

Specifically, since multiple threads can be trying to check the agents at the same time, even though the current setup stops them from reconfiguring the same agent in parallel, they can still reconfigure the same agent sequentially, because actually checking the old config/deps vs. the new ones is not done within the locked portion.

We really don't need multiple threads trying to check the agents at once. It would be better if the lock was at a bit of a higher level. This also means less time spent trying to acquire locks (one per the process of checking the agents vs. one per each agent we check).