Kubernetes Scheduling Explained for Developers

🛑 The Kubernetes scheduler doesn't randomly assign pods to nodes. It runs a two-phase pipeline for every pod: 1. Filtering — discard every node that can't run the pod (wrong labels, not enough resources, taints) 2. Scoring — rank what's left from 0 to 100 across 13 plugins. Highest score wins. One thing most people miss: resource requests. If you don't set them, the scheduler treats your pod as having a size of 0. It packs nodes until the kubelet starts evicting pods to reclaim memory. Another one: topology spread constraints. They sound great — "spread my replicas across zones." But if one zone loses capacity, the replacement pod can't go to other zones without violating the skew. It stays Pending. I wrote a guide that walks through the scheduler internals with 30+ diagrams: - The scheduling cycle (queue → filter → score → bind) - nodeSelector, affinity, anti-affinity, taints - Topology spread constraints and their failure modes - Requests vs actual usage - Priority classes and preemption - The scheduler simulator for testing constraints locally 👉 https://lnkd.in/g3YNGwRb 📨 And if you like content like this, you might like the Kube Today newsletter I curate at https://lnkd.in/gF-ctXXJ

Most people assume the kube-scheduler has a central decision engine that decides where pods go. Well, it doesn’t. The real intelligence lies in its Plugin Framework. Every operation, resource checks, node filtering, scoring, and even binding, is implemented as a plugin. The scheduler core itself orchestrates the execution of those plugins, manages state, and keeps the system consistent. This plugin-first architecture is what makes Kubernetes scheduling so extensible and maintainable. It’s why features like asynchronous preemption and #DRA can integrate without rewriting the core. In my latest deep dive, I break down: - How the Plugin Framework works - The scheduler’s supporting infrastructure (queues, cache, events) - The synchronous and asynchronous cycles - And how everything fits together into a high-performance orchestration engine If you’ve ever wanted to understand the scheduler beyond the surface or plan to write custom scheduling logic, this series is for you. Read Part 1: Inside kube-scheduler: The Plugin Framework That Powers Kubernetes Scheduling 👉 https://lnkd.in/g8VzD38H #Kubernetes #CloudNative #Scheduler #SIGScheduling #OpenSource #SoftwareArchitecture