A few thoughts:

• We need a way to tell the broker to shut down. This may be through the monitor, or mabye a new "admin" connection type.
• The broker needs to tell the workers to shut down cleanly, and then exit cleanly itself.
• We should consider what happens when the broker sends the message and then immediately exits. (Maybe this isn't an issue).
• We should consider what happens when the worker takes a long time to start and the broker finishes before the worker connects. In this situation, we would ideally have the worker exit cleanly.
• When the worker gets a shutdown message from the broker, it should cancel all outstanding jobs like now, but then exit with a success exit code.