medium10 pts · first answer countsmonitoringlatency

Fleet-wide p99 from many pods

Each of 20 pods exports its own p99 latency. A dashboard shows the fleet p99 as the average of the 20 values. What is the problem, and what is the fix?

A.Averaging percentiles does not give a real percentile; aggregate histogram buckets across pods and compute the quantile from themB.Nothing; the average of p99 values is the fleet p99C.Use the minimum of the 20 values insteadD.Percentiles cannot be measured on Kubernetes

← What to log for each LLM call Online features stopped updating →