Skip to content

Opinion & Overview

These are practical, opinionated operator guides for decisions that are bigger than daily update cycles. Use this section for architecture tradeoffs, reliability and resilience patterns, and clear point-of-view guidance for platform teams.

Date Article Summary
2026-04-01 Node Pressure and Eviction: What Kubelet Actually Does When Things Get Tight Kubelet's eviction manager is the last line of defense before a node becomes unstable. Understanding how eviction thresholds work, why QoS class determines who dies first, and where the gap between scheduler requests and…
2026-03-17 Horizontal Pod Autoscaler: From Metrics to Scheduling Pressure HPA is a proportional feedback controller, not a provisioning system. Understanding its formula, metrics pipeline lag, and stabilization layer is what separates operators who tune it reliably from those who chase…
2026-03-16 How etcd Consistency Guarantees Shape Kubernetes Control Plane Behavior etcd is not just a database. It is the linearizable coordination primitive that every Kubernetes controller is built on top of.
2026-03-14 The Kubernetes Scheduler: Decision Loop, Plugin Architecture, and Operational Reality The Kubernetes scheduler is not a router. It is a continuous reconciler that applies a filtering and ranking pipeline to a cluster snapshot.
2026-03-12 Why Every Pod Gets a Real IP: The Kubernetes Flat Network Model, CNIs, and Ingress Controllers Kubernetes enforces a simple rule: every Pod can reach every other Pod directly, without NAT. This single constraint drives the entire networking architecture - from CNI plugin selection to ingress controller design.
2026-03-11 How Traffic Actually Flows in Kubernetes: Services, kube-proxy, and Cloud Load Balancers A Kubernetes Service is not a load balancer - it is a routing abstraction. Understanding the three layers that actually move traffic (node dataplane, cloud load balancer, application connections) explains most real-world…
2026-03-10 True HA in Kubernetes: Why Multi-Zone Alone Isn't Enough Multi-zone clusters remove the zone-failure risk but create a false sense of security. Real high availability requires coordinating PDBs, topology spread constraints, affinity rules, probes, graceful shutdown, and…
2026-03-10 Why Kubernetes Scheduling Uses Requests, Not Limits Kubernetes schedules Pods based on resource requests, not limits. Understanding this distinction explains noisy neighbor problems, unexpected autoscaling behavior, and wasted cluster capacity, and how to fix them.