Scaling

Platformatic ICC supports autoscaling applications on both Kubernetes and AWS ECS. It adjusts the number of application instances based on Node.js runtime signals and configured safety limits.

Application instances

An application instance is one independently scaled copy of a Watt application—for example, a Kubernetes pod or an AWS ECS task. Throughout this section, instance refers to this application copy, rather than an underlying compute host or a worker running inside Watt.

ICC provides two scaler versions, selected for the whole ICC deployment with PLT_SCALER_ALGORITHM_VERSION.

Version	Availability	Behavior
Reactive scaler (V1)	Open Source and Enterprise	Reacts to current Event Loop Utilization (ELU) and heap pressure.
Predictive scaler (V2)	Enterprise	Forecasts short-term ELU and heap load and provisions capacity before the forecast crosses a threshold.

The two versions are alternatives: ICC runs either V1 or V2. They do not run together.

Enterprise deployments using V2 also include the Planner. The Planner learns recurring capacity requirements and lets operators schedule minimum instance counts before expected demand. It complements the live V2 scaler; it is not a third live-scaling algorithm.

How scaling fits into ICC

1. @platformatic/watt-extra collects application-level runtime signals from each Watt instance.
2. The selected scaler evaluates those signals and calculates a target instance count.
3. ICC applies minimum and maximum instance limits.
4. The Machinist service updates the application’s desired instance count through the infrastructure provider.
5. ICC records scaling events and exposes their reasons and metrics in the dashboard.

The following diagram shows the high-level real-time scaling data flow on Kubernetes. Each Watt pod sends runtime metrics through Watt Extra, and ICC applies the resulting replica target to Kubernetes.

Select the scaler version

V1 is the default:

PLT_SCALER_ALGORITHM_VERSION=v1

Enterprise deployments can select V2:

PLT_SCALER_ALGORITHM_VERSION=v2

Caution

V2 is an Enterprise feature. An Open Source ICC deployment configured with v2 will fail to initialize the predictive scaler.

Restart ICC after changing the algorithm version. Applications must meet the Watt Extra and Platformatic Runtime version requirements.

Next steps

• Verify the scaling prerequisites.
• Configure scaling limits for each application.
• Learn how the Reactive scaler (V1) makes decisions.
• Learn how the Predictive scaler (V2) forecasts load.
• Configure recurring capacity with the Planner.
• Review all scaler configuration variables.