Kubernetes Node Resource Management

10/13/2024

This post introduces a practical approach to diagnosing and fixing Kubernetes node resource exhaustion. It provides specific `kubectl` commands for identifying nodes with high memory usage and for inspecting individual node resource allocation. The core contribution is the methodology for pinpointing problematic pods, adjusting their resource requests/limits, and ensuring correct node placement via `nodeSelector`. It also offers cleanup commands for pods, services, and deployments.

11/13/2019

This post details the architectural shift in Helm 3 from a client-server model (Helm 2 with Tiller) to a client-only architecture. It explains the removal of Tiller, the implications for security (delegation to Kubernetes user cluster security), and the changes in release storage (Secrets in release namespace, new naming convention). It also covers modifications to charts, including the consolidation of dependency declarations in Chart.yaml and the introduction of library charts. Other changes like XDG base directory specification support and simplified CRD handling are also discussed.