Docker and CI/CD · Senior · Question 27 of 75
Why can CPU-based autoscaling work poorly for an I/O-bound Node service?
- 1Scaling on CPU always removes pods while load is still rising
- ✓Correct answer: It can be saturated on connections or event-loop lag at low CPU
- 3Kubernetes can't measure CPU for containers running Node.js
Explanation
If the bottleneck is waiting on I/O, CPU stays low while latency grows. Scale on request rate, concurrency, queue depth or latency via custom metrics.