Workload Autoscaler

2 min read

The Workload Autoscaler scales the Deployments your operator manages based on time, external queue depth, sibling operator metrics, or any combination. It is declared directly on the Deployment:

onCreate:
  deployments:
    - name: "{{ .metadata.name }}"
      replicas: "{{ .spec.replicas }}"
      autoscale:
        min: 2
        max: 20
        cooldown: 3m
        scaleUp:
          conditions:
            when:
              - field: external.queue.pendingJobs
                greaterThan: "100"
          increment: 2
        scaleDown:
          conditions:
            when:
              - field: external.queue.pendingJobs
                lessThan: "10"
          decrement: 1

What the workload autoscaler does

  • Evaluates scaleUp and scaleDown conditions on every reconcile cycle
  • Patches spec.replicas on the Deployment when conditions are met
  • Enforces min and max boundaries
  • Respects a cooldown period between scale events to prevent oscillation
  • Works with any source the Resolver reads — external:, cross:, time notes, user-defined notes

Relationship to the Operator Autoscaler

The Operator Autoscaler (operatorBox.autoscale) scales the operator’s own machinery: workers, queue depth limit, resync period. It answers: how fast should the operator reconcile?

The Workload Autoscaler (deployments[].autoscale) scales the resources the operator manages: Deployment replicas. It answers: how many instances of the managed workload should run?

They share the same condition engine and run in the same reconcile loop. They are independent — you can use one, both, or neither.

Documentation structure

1. Scaling Signals

Time conditions, external HTTP endpoints, cross-operator metrics, notes, and composite conditions — every source the Resolver reads is a scaling signal.

2. Schema Reference

Scaling modes (target vs increment/decrement), the full field reference, and validation rules.

Try it

ork init --pack use-cases/workload-autoscaler
# Follow the steps in the README