Reading series
Production Systems
4 parts, in reading order.
distributed systemsThe Acknowledgment Gap - How Event-Driven Systems Lose Messages Without Errors
Event-driven systems can drop messages while every component reports success. How acknowledgment boundaries hide loss, and where to put them.
distributed systemsWhen "no healthy upstream" isn't about the upstream you think
Learn what Envoy's no healthy upstream error means, why health checks can remove every backend, and how timeouts and retries can trigger it.
distributed systemsReplacing a Critical Data Path Without a Flag Day
Shadow reads, a normalized comparison and a staged cutover: how to replace a data path in production without a single big-bang release.
distributed systemsReplacing Sharp with WASM: How a Hidden SVG Dependency Broke Our Rendering Pipeline
A production-only SVG failure exposed a hidden Sharp dependency, a broken fallback, and why I replaced the rasteriser with resvg WASM.