Why did an API key management platform abandon edge serverless for stateful containers? Unkey hit 30ms p99 cache latency when they needed sub-10ms—so they rebuilt everything on AWS Fargate. This episode covers the technical decision-making framework for choosing between serverless and containers, plus a deep dive into Kubernetes 1.35's new structured z-pages for debugging. In This Episode:- The serverless constraint: stateless = network request for every cache read- Unkey's complexity tax: Workers, Durable Objects, Queues, custom proxies- The container solution: Fargate + Global Accelerator = 6x performance- Decision framework: latency targets, data hotness, complexity budget- K8s 1.35 z-pages: JSON structured responses for compliance automation Key Statistics:- 30ms p99 cache latency before migration (target: <10ms)- 6x performance improvement after moving to containers- Self-hosting unlocked as unexpected bonus New episodes drop weekly. Subscribe to stay current on platform engineering. Links:Full show notes: https://platformengineeringplaybook.com/podcasts/00077-unkey-serverless-containers-migrationContribute: Open a PR on GitHub