Previously, when the state server received a request to globally lock a state key, it would block until it acquired that lock. This can result in the following deadlock scenario:

• We have s server threads serving requests.
• s + 1 clients request a lock on a state key.
• One of them gets the lock, the other s cause all the server threads to block.
• The one with the lock tries to request an unlock, but all the server threads are blocking on the lock.

Rather than fix this, I've decided to remove global state locks for the following reasons:

• Global locks are implemented without this problem in the distributed synchronisation primitives (as part of the point-to-point group operations).
• There are ways to avoid needing a global lock entirely, such as using the state append operation to queue updates, or collective communication operations like reduce.

As part of debugging this issue I also tidied up the logging in the state server and client a bit.