Back to all scenarios
Scenario #402
Scaling & Load
Kubernetes v1.24, EKS, Burstable QoS

CPU Throttling Prevented Effective Autoscaling

Application CPU throttled even under low usage, leading to delayed scaling.

Find this helpful?
What Happened

HPA didn’t trigger scale-up due to misleading low CPU usage stats.

Diagnosis Steps
  • 1Metrics showed low CPU, but app performance was poor.
  • 2kubectl top pod confirmed low utilization.
  • 3cgroups showed heavy throttling.
Root Cause

CPU limits were set too close to requests, causing throttling.

Fix/Workaround
• Increased CPU limits or removed them entirely for key services.
Lessons Learned

CPU throttling can suppress scaling metrics.

How to Avoid
  • 1Monitor cgroup throttling stats.
  • 2Tune CPU requests/limits carefully.
Previous ScenarioNext Scenario