Scalingo web container autoscaling metric target and cooldown configuration

Goal is to configure autoscaling for a Scalingo web container with a defined min and max container count driven by a metric such as RPM or resource usage.

The configuration requires choosing a target value and an aggregation window for the metric, plus bounds to limit cost and capacity. Statelessness is assumed for horizontal scaling, and database connection limits are a separate constraint outside the autoscaler.

Autoscaling decisions depend on target value and aggregation window, and misconfigured values can lead to oscillation or slow reaction to traffic changes. The exact semantics for target interpretation and cooldown behavior are platform-dependent.

What aggregation window is applied to the chosen metric when evaluating the target? Is the target interpreted as per-container average or total across containers? Are scale-up and scale-down cooldowns configurable separately and what are the defaults?