Documentation
Docs
Agent Development

Scaling Configuration

Configure Horizontal Pod Autoscaling (HPA) and KEDA triggers.

Auto-Scaling Strategy

Consonant supports scaling based on CPU, Memory, or Custom Metrics (via KEDA).

CPU Scaling (Simple)

The easiest way to scale is by CPU utilization.

yaml
scale:
  minReplicas: 1
  maxReplicas: 10
  cpuUtilization: 80

Event-Driven Scaling (KEDA)

For queue-based workers, use KEDA triggers.

yaml
scale:
  minReplicas: 0
  maxReplicas: 50
  triggers:
    - type: rabbitmq
      metadata:
        queueName: agent-tasks
        queueLength: "5"
Scale to Zero
Setting minReplicas: 0 allows your agents to shut down completely when idle, saving huge costs.