Schema Reference

1 min read

Scaling modes

Two modes, mutually exclusive per direction:

target — jump scaling. When conditions are true, set replicas to exactly this value. Appropriate for time-based scaling where the desired state is known. Converges in one reconcile cycle.

increment / decrement — step scaling. Add or remove N replicas per evaluation. Appropriate for metric-driven scaling where gradual ramp-up avoids overshooting. Each reconcile that finds conditions true adds one step.


Field reference

FieldRequiredDescription
maxyesCeiling — autoscaler never exceeds this value
minnoFloor — defaults to the resolved replicas value
cooldownnoWait between scale events (e.g. 3m, 30s). Default 1m
scaleUp.conditionsnowhen: / anyOf: conditions that trigger scale-up
scaleUp.targetnoJump to this exact replica count
scaleUp.incrementnoAdd this many replicas per evaluation tick
scaleDown.conditionsnoConditions that trigger scale-down
scaleDown.targetnoJump to this exact replica count
scaleDown.decrementnoRemove this many replicas per evaluation tick

Validation

ork validate enforces at load time:

  • max is required — a workload autoscaler with no ceiling fails validation
  • min < max — if both are declared and min >= max, validation fails
  • target and increment/decrement are mutually exclusive per direction — if both are declared in the same scaleUp or scaleDown block, validation fails