Scaling Configuration

This page lists the Scaler service configuration variables. Defaults reflect the current ICC implementation.

Caution

The V2 prediction parameters form a coordinated model. Change advanced values only after testing them against representative traffic, instance startup behavior, and failure conditions.

Common configuration

These settings apply to both scaler versions.

Variable	Default	Description
PLT_SCALER_ALGORITHM_VERSION	v1	Selects v1 reactive scaling or v2 predictive scaling. V2 is Enterprise-only.
PLT_SCALER_SCALING_DISABLED	false	Prevents the scaler from changing controller replicas.
PLT_SCALER_MIN_PODS_DEFAULT	1	Default application minimum when no stored limit exists.
PLT_SCALER_MAX_PODS_DEFAULT	10	Default application maximum when no stored limit exists.
PLT_SCALER_SYNC_K8S	60	Controller and label synchronization interval, in seconds.
PLT_SCALER_POD_MIN_LABEL	icc.platformatic.dev/scaler-min	Controller label containing the application minimum.
PLT_SCALER_POD_MIN_DEFAULT_VALUE	1	Minimum used during label synchronization when only the maximum label is provided and no stored minimum exists.
PLT_SCALER_POD_MAX_LABEL	icc.platformatic.dev/scaler-max	Controller label containing the application maximum.
PLT_SCALER_POD_MAX_DEFAULT_VALUE	10	Maximum used during label synchronization when only the minimum label is provided and no stored maximum exists.

See Scaling Limits for application-level configuration.

Service dependencies

Variable	Default	Description
PLT_ICC_VALKEY_CONNECTION_STRING	Required	Valkey connection used for scaler state, cooldowns, signals, and scheduled limits.
PLT_MACHINIST_URL	Required	Machinist service used to read and update application controllers.
PLT_MACHINIST_PROVIDER	k8s	Infrastructure provider prefix used by the Machinist client.
PLT_METRICS_PROMETHEUS_URL	None	Prometheus endpoint used by V1 and scaling-event metric views.

Reactive scaler (V1)

Variable	Default	Description
PLT_METRICS_TIME_RANGE	60	Prometheus history window used for application metrics, in seconds.
PLT_SCALER_DEBOUNCE	10000	Alert debounce interval, in milliseconds.
PLT_SCALER_MAX_HISTORY	10	Maximum recent scaling events retained per application.
PLT_SCALER_MAX_CLUSTERS	5	Maximum performance-history clusters retained per application.
PLT_SCALER_ELU_THRESHOLD	0.8	ELU threshold used by the current V1 executor.
PLT_SCALER_HEAP_THRESHOLD	0.85	Heap-utilization threshold.
PLT_SCALER_POST_EVAL_WINDOW	300	Delay before evaluating a scaling result, in seconds.
PLT_SCALER_COOLDOWN	15	Scale-up cooldown, in seconds. The scale-down cooldown is derived from this value.
PLT_SCALER_PERIODIC_TRIGGER	60	Periodic metrics evaluation interval, in seconds.

Predictive scaler (V2)

The following settings are used only when PLT_SCALER_ALGORITHM_VERSION=v2.

Prediction lifecycle

Variable	Default	Description
PLT_SCALER_DASHBOARD_API_ENABLED	false	Includes additional prediction-debug fields in the dashboard API.
PLT_SIGNALS_SCALER_RECONNECT_TIMEOUT_MS	5000	Grace period before treating a disconnected runtime as terminated.
PLT_SIGNALS_SCALER_READINESS_DELAY_MS	10000	Readiness-probe interval allowance used when estimating when traffic reaches a new instance.
PLT_SIGNALS_SCALER_HORIZONTAL_TREND_THRESHOLD	0.2	Normalized trend angle, in degrees, below which a trend is considered horizontal.
PLT_SIGNALS_SCALER_PENDING_SCALE_UP_EXPIRY_MS	60000	Lifetime of a pending scale-up record.
PLT_SIGNALS_SCALER_REDEPLOY_TIMEOUT_MS	120000	Window used to recognize a deployment replacement rather than a scaler action.
PLT_SIGNALS_SCALER_INIT_TIMEOUT_MS	30000	Initial instance startup-time estimate.
PLT_SIGNALS_SCALER_HORIZON_MULTIPLIER	1.2	Safety multiplier applied to the learned startup time.
PLT_SIGNALS_SCALER_MIN_HORIZON_MS	25000	Minimum prediction horizon.
PLT_SIGNALS_SCALER_MAX_HORIZON_MS	40000	Maximum prediction horizon.

Startup-time learning

Variable	Default	Description
PLT_SIGNALS_SCALER_INIT_TIMEOUT_WINDOW_SIZE	5	Number of recent observed startup times used for the median target.
PLT_SIGNALS_SCALER_INIT_TIMEOUT_STEP_RATE	0.1	Maximum proportional change applied in one update.
PLT_SIGNALS_SCALER_INIT_TIMEOUT_UP_FACTOR	1.5	Multiplier allowing startup estimates to increase faster.
PLT_SIGNALS_SCALER_INIT_TIMEOUT_DOWN_FACTOR	1.0	Multiplier controlling decreases in the startup estimate.

Scaling decisions

Variable	Default	Description
PLT_SIGNALS_SCALER_SCALE_UP_K	1.5	Dampening coefficient applied to forecast growth during scale-up calculation.
PLT_SIGNALS_SCALER_SCALE_UP_MARGIN	0.1	Additional scale-up margin used when forecast capacity is slightly over an instance boundary.
PLT_SIGNALS_SCALER_SCALE_DOWN_MARGIN	0.3	Spare-capacity margin retained during scale-down.
PLT_SIGNALS_SCALER_LIVE_METRIC_COUNT	3	Most recent raw samples per worker used for the live metric.
PLT_SIGNALS_SCALER_MAX_SCALE_UP_STEP	4	Maximum instances added by a single V2 decision.
PLT_SIGNALS_SCALER_PROCESSING_INIT_TIMEOUT_MS	1000	Delay before processing the first pending signal batch.
PLT_SIGNALS_SCALER_PROCESSING_COOLDOWN_MS	10000	Delay before processing another pending batch after a decision cycle.
PLT_SIGNALS_SCALER_INSTANCES_WINDOW_MS	180000	Runtime lifecycle history retained for reconstruction and deployment detection.

Directional cooldowns

Variable	Default	Description
PLT_SIGNALS_SCALER_COOLDOWN_SCALE_UP_AFTER_SCALE_UP_MS	5000	Scale-up delay after a scale-up.
PLT_SIGNALS_SCALER_COOLDOWN_SCALE_UP_AFTER_SCALE_DOWN_MS	5000	Scale-up delay after a scale-down.
PLT_SIGNALS_SCALER_COOLDOWN_SCALE_DOWN_AFTER_SCALE_UP_MS	180000	Scale-down delay after a scale-up.
PLT_SIGNALS_SCALER_COOLDOWN_SCALE_DOWN_AFTER_SCALE_DOWN_MS	60000	Scale-down delay after a scale-down.

ELU forecast

Variable	Default	Description
PLT_SIGNALS_SCALER_ELU_WINDOW_MS	90000	ELU history window used for reconstruction and forecasting.
PLT_SIGNALS_SCALER_ELU_SAMPLE_INTERVAL	1000	ELU alignment interval.
PLT_SIGNALS_SCALER_ELU_REDISTRIBUTION_MS	30000	Period over which a new instance’s ELU contribution is introduced.
PLT_SIGNALS_SCALER_ELU_ALPHA_UP	0.2	Holt level smoothing factor when ELU is above the forecast.
PLT_SIGNALS_SCALER_ELU_ALPHA_DOWN	0.1	Holt level smoothing factor when ELU is below the forecast.
PLT_SIGNALS_SCALER_ELU_BETA_UP	0.12	Holt trend smoothing factor when ELU is above the forecast.
PLT_SIGNALS_SCALER_ELU_BETA_DOWN	0.1	Holt trend smoothing factor when ELU is below the forecast.

Heap forecast

Variable	Default	Description
PLT_SIGNALS_SCALER_HEAP_WINDOW_MS	90000	Heap history window used for reconstruction and forecasting.
PLT_SIGNALS_SCALER_HEAP_SAMPLE_INTERVAL	1000	Heap alignment interval.
PLT_SIGNALS_SCALER_HEAP_REDISTRIBUTION_MS	30000	Period over which a new instance’s heap contribution is introduced.
PLT_SIGNALS_SCALER_HEAP_ALPHA_UP	0.2	Holt level smoothing factor when heap is above the forecast.
PLT_SIGNALS_SCALER_HEAP_ALPHA_DOWN	0.1	Holt level smoothing factor when heap is below the forecast.
PLT_SIGNALS_SCALER_HEAP_BETA_UP	0.12	Holt trend smoothing factor when heap is above the forecast.
PLT_SIGNALS_SCALER_HEAP_BETA_DOWN	0.1	Holt trend smoothing factor when heap is below the forecast.
PLT_SIGNALS_SCALER_VALKEY_KEY_PREFIX	scaler:	Prefix for V2 state stored in Valkey.

V2 thresholds are supplied with the per-service runtime signals rather than configured through a global Scaler environment variable.

Planner and scheduler

Planner settings apply only to V2 Enterprise deployments.

Variable	Default	Description
PLT_SCALER_SCHEDULER_ENABLED	true	Enables application of accepted suggestions and custom schedules.
PLT_SCALER_SCHEDULER_TICK_MINUTES	1	Interval for evaluating active scheduled windows.
PLT_SCALER_TIME_SLOT_MINUTES	5	Collection interval for raw required/actual instance statistics. Must divide 60 evenly.
PLT_SCALER_TIME_SLOT_TIMEZONE	UTC	Cluster calendar timezone used for history, predictions, and plan rebuilding.
PLT_SCALER_TIME_WINDOW_MINUTES	60	Planner aggregation and scheduling window. Must be a multiple of the slot and divide 1440.
PLT_SCALER_PATTERN_PERCENTILE	p50	Time-weighted percentile selected from each slot for pattern learning.
PLT_SCALER_PATTERN_PREDICTION_DAYS	60	Number of future days generated by the Planner.
PLT_SCALER_SUGGESTION_OCCURRENCES_LIMIT	14	Maximum past occurrences shown as evidence for a suggestion.
PLT_SCALER_WINDOW_CATEGORIES	5	Number of relative capacity categories displayed in the Planner. Must be an odd integer of at least 3.
PLT_SCALER_WINDOW_MIN_PODS	3	Minimum category width in instances.
PLT_SCALER_WINDOW_MIN_FRACTION	0.1	Minimum category width as a fraction of the application’s capacity range.