Serverless

Serverless platforms (Lambda, Cloud Functions, Cloud Run, Azure Functions, Workers) run your code without managing servers — billing often ties to invocations, duration, and memory. FullStack JS teams use them for webhooks, cron, image thumbnailing, and BFF endpoints while keeping long-lived WebSockets or heavy CPU on containers or VMs.

| Factor | Implication | |--------|-------------| | Cold start | Keep bundles small; provisioned concurrency if needed | | Timeout | Split long jobs to queues | | State | Externalize to DB, cache, or object storage | | Observability | Structured logs, traces, cold-start metrics |

Package with esbuild or similar; avoid shipping entire `node_modules` when a minimal handler suffices. Use environment-specific config and secrets managers, not committed env files.

On interviews: when serverless beats containers, cold-start mitigation, VPC-attached Lambda trade-offs, local dev parity, and vendor limits (payload size, concurrency).

Common pitfalls: 15-minute batch jobs in functions; connection per invocation to Postgres; giant dependencies; assuming full Node API surface on edge runtimes.

The trade-off is ops simplicity and elastic scale versus unpredictable latency, execution limits, and platform coupling.

Checklist:

• Right-size memory after profiling duration cost.
• Use queues for work exceeding timeout budgets.
• Share nothing; externalize session and cache state.
• Measure cold starts in staging with realistic bundles.