Why Kubernetes Pod Disruption Budgets Fail Under Spot Node Drains

Episode notes

Episode 60 of DevOps Daily with Fexingo: Lucas and Luna dig into a common but poorly understood Kubernetes failure mode — Pod Disruption Budgets (PDBs) that actually accelerate downtime during spot instance preemptions. Using AWS Spot Node draining as the concrete case (similar dynamics apply on GKE Preemptible and Azure Spot), they trace why PDBs designed to protect availability can paradoxically block node drains, leaving workloads stranded. They walk through a real scenario: a stateless microservice with minAvailable=2, three replicas, and a single-node spot pool. The drain command hangs because Kubernetes can't find a second node to reschedule to, triggering a timeout instead of a graceful eviction. They discuss the root cause — PDBs don't consider node topology — and offer two practical mitigations: setting maxUnavailable instead of minAvailable where possible, and combining PDBs with cluster autoscaler priority classes. No theory, no fluff — just a production pitfall you'll face if you use spot instances at scale.