An 8‑year ops veteran investigates a night‑time alert showing 900% CPU usage, discovers that a JVM inside a Kubernetes pod misreads host cores while the container is limited to two CPUs, and outlines how improper thread‑pool settings and monitoring metrics caused massive throttling before presenting concrete fixes.
This article explains why memory OOM and CPU throttling are critical issues in Kubernetes, shows how limits and requests work, demonstrates monitoring techniques with Prometheus and cAdvisor, and provides practical best‑practice recommendations to avoid pod eviction and performance degradation.
Uber reduced cluster-wide CPU allocation by 11% and stabilized P99 latency by replacing CPU quota throttling with cpusets, a CPU‑pinning technique that isolates containers on specific cores, eliminating throttling while only slightly increasing P50 latency.
This article explains how Kubernetes resource limits function, how to interpret CPU limits as time slices, the Linux accounting system behind them, relevant Prometheus metrics for detecting throttling, practical examples with multithreaded containers, and guidance on setting alerts and avoiding performance pitfalls.
This article explains why Kubernetes pods encounter out‑of‑memory errors and CPU throttling, how limits and requests influence resource allocation, and provides practical monitoring techniques using Prometheus and cAdvisor to proactively identify and mitigate these issues before they impact performance or cause pod eviction.
The article explains how Uber replaced CPU quota enforcement with cpusets (CPU pinning) for stateful workloads, reducing P99 latency, improving performance consistency, and saving about 11% of cores across the cluster by eliminating throttling.
This article explains why container workloads suffer from CPU throttling and NUMA‑related performance loss in cloud‑native environments, examines Kubelet's CPU allocation policies, demonstrates the impact of CPU bursts and topology‑aware scheduling, and shows how Alibaba Cloud ACK mitigates these issues with concrete data.
Running Java services on Kubernetes often suffers from long JVM warm‑up times that cause high latency during deployments, but by analyzing CPU throttling and leveraging Burstable QoS with appropriate request/limit settings, you can dramatically reduce warm‑up delays without extra pods or cost.
