Back to all scenarios
Scenario #403
Scaling & Load
Kubernetes v1.21, on-prem

Overprovisioned Pods Starved the Cluster

Aggressively overprovisioned pod resources led to failed scheduling and throttling.

Find this helpful?
What Happened

Apps were deployed with excessive CPU/memory, blocking HPA and new workloads.

Diagnosis Steps
  • 1kubectl describe node: Insufficient CPU errors.
  • 2Top nodes showed 50% actual usage, 100% requested.
Root Cause

Reserved resources were never used but blocked the scheduler.

Fix/Workaround
• Adjusted requests/limits based on real usage.
Lessons Learned

Resource requests ≠ real consumption.

How to Avoid
  • 1Right-size pods using VPA recommendations or Prometheus usage data.
Previous ScenarioNext Scenario