Question

An image-thumbnailing service (Go) runs on Kubernetes. After a routine traffic increase (a marketing push doubled uploads), p99 latency degraded badly and requests started timing out, but per-pod CPU utilization shown in your dashboard reads only ~55% — nowhere near 100%. The HPA isn't scaling up because it's targeting 70% CPU and 55 < 70. The pod has `resources.limits.cpu: 1` and `requests.cpu: 0.5`. The container metric `container_cpu_cfs_throttled_periods` is high and rising. Triage and mitigate.