There have been a few instances where a worker's internal state has been discovered to be inconsistent. This might be a validation error or an invalid transition or some other internal error. We should resolve these and determine what is going on. In the meantime it's also critical that we provide a stable experience to users. These situations are rare enough that it probably makes sense to just close down the worker and have it be restarted, and hope that the scheduler logic (which appears to be in a better state today) cleans things up properly.

This doesn't solve the underlying problem, but does give us a little space and gives users a better experience while we discover and resolve that problem.