I need better metrics to auto-scale machine runners effectively. Currently, our auto-scaling is time-based, and we need it to be on-demand based on queue times and other metrics.