Back to all scenarios
Scenario #407
Scaling & Load
Kubernetes v1.25, KEDA + Knative

Scale-To-Zero Caused Cold Starts and SLA Violations

Pods scaled to zero, but requests during cold start breached SLA.

Find this helpful?
What Happened

First request after inactivity hit cold-start delay of ~15s.

Diagnosis Steps
  • 1Prometheus response latency showed spikes after idle periods.
  • 2Knative logs: cold-start events.
Root Cause

Cold starts on scale-from-zero under high latency constraint.

Fix/Workaround
• Added minReplicaCount: 1 to high-SLA services.
Lessons Learned

Scale-to-zero saves cost, but not for latency-sensitive apps.

How to Avoid
  • 1Use minReplicaCount and warmers for performance-critical services.
Previous ScenarioNext Scenario