This article explains how a Java application's memory usage can cause Kubernetes pod OOM, the role of namespaces and cgroups in container isolation and resource limiting, and practical steps to adjust JVM settings to prevent frequent restarts.
This article explains the memory challenges of containers in cloud‑native Kubernetes environments, describes how Kubernetes maps Request/Limit to cgroups, outlines the limitations of overcommit and mixed workloads, and details Alibaba Cloud Linux 2's Memory QoS features—including memory.min, memory.high, async reclaim, and global water‑mark adjustments—that improve fairness and latency, supported by Redis benchmark results.
After moving IoT services to Kubernetes, containers were OOM‑killed despite RSS staying below limits because Kubernetes bases OOM decisions on the Working Set Size (WSS) metric, which includes file cache, and the article explains its calculation, reproduces the issue, and offers practical mitigation strategies.
