At the moment, the workerRegistrationDuration metric captures the time between "worker requested" and "registerWorker being called". This is a good start, but by itself, it conflates time that we control (eg: VM start-up time, start-worker/generic-worker start-up time) with things outside of our control (eg: VM provisioning time). It would be great to get augmented or additional metrics here to increase the granularity here, to help inform whether or not we have something actionable when we see spikes.