hard20 pts · first answer countskubernetesperformanceresource-management

Throttled at 30 percent CPU

A multi-threaded API with a 1-CPU limit shows p99 latency spikes. Average CPU use is only 0.3 cores, yet container_cpu_cfs_throttled_periods_total climbs steadily. What explains this?

A.The metric is misleading; the kernel cannot throttle a container below its CPU limit, so it is a cAdvisor bug.B.Memory pressure on the node is counted by cAdvisor as CPU throttling, so the real cause is RAM.C.CFS quota is per 100 ms period; bursts of threads use it up quickly and then wait, which averages hide.D.The CPU request is set higher than the limit, which the kubelet then reports as throttling.

← Risky GitHub Actions trigger Slow external DNS in Pods →