Currently autoscaler evaluate the metrics in a breach duration, and it may be slow to do scale action when there are suddenly many requests come to the app.
So we can set 2 breach durations, the stale one and the panic one. For example, stable duration is 60s and panic duration is 6s. The autoscaler does evaluation for both stable and panic duration, and if the one for panic has been breached twice, then it use the panic duration. And if there is no scale action for some time, it then goes back to stable duration.