Cold Start

Cold starts happen when a serverless function or container has been idle and the host needs to spin up a new instance for the next request. Latency varies by host and runtime: Vercel serverless Lambda cold start is 300-1500 ms. Render free web service spin-up is 30+ seconds. Cloudflare Workers cold start is under 5 ms (V8 isolates wake up nearly instantly). For an indie product, cold starts decide whether your free tier feels acceptable or feels broken. Render's 15-minute spin-down on free web services is the classic case: visitors who arrive between 9 PM and 9 AM wait 30 seconds for the page to start loading. Most click away. Mitigations: warming pings (cron hits your endpoint every 5 minutes), keeping a function warm via concurrent invocations, or moving to a runtime that doesn't have meaningful cold starts (Cloudflare Workers, Deno Deploy) for latency-sensitive paths. Edge functions on V8 isolates effectively have no cold start. Lambda-style serverless does. Container-based hosts (Render, Fly, Railway) have the worst cold start because the whole container needs to boot, not just a runtime.