Skip to content

Best Practices

Use these guides to configure production behavior and verify it under load.

GuideUse it to
Kubernetes Autoscaling Best PracticesSet requests, probes, placement and shutdown behavior
KEDA Best PracticesTune activation, fallback, polling and stabilization
Scale-to-zero & cold startsBalance idle capacity against startup latency
  1. Set an acceptance target: peak throughput, latency, error rate and recovery time.
  2. Check ownership: one controller per replica/resource field; configure generated HPA/KPA resources through the ScaledObject.
  3. Test ramp, spike and idle traffic: compare source metrics, desired replicas and ready capacity, including node provisioning and startup.
  4. Test failure and scale-down: unavailable metrics, controller failover, request draining and queue acknowledgements.
  5. Retain a rollback: record the original configuration and apply changes through the same Helm/GitOps owner.

Autoscaling Checks exercise the signal-to-replica path and expose check latency for alerting. They complement application load tests. Use Monitor scaling to locate delays.

For component operations, see high availability and performance tuning in Operate. See also resource sizing, HTTP proxy tuning and rightsizing validation.