serverless-architecture

Serverless fits spiky, unpredictable, low-baseline, or event-driven workloads where scale-to-zero eliminates idle cost. Steady high-throughput or latency-critical services are cheaper and more predictable on always-on containers or VMs.

Trim package size and dependencies, move expensive initialization to module scope so warm invocations reuse it, choose lightweight runtimes, and avoid VPC attachment unless needed. For latency-critical paths, use provisioned concurrency.

No. Synchronous function-to-function chains bill the caller while it waits and couple timeouts. Use an orchestrator like AWS Step Functions for multi-step flows, or decouple steps with a queue or pub/sub topic.

Queues and event sources guarantee at-least-once delivery, so handlers can run multiple times on the same message after retries or duplicates. Idempotent handlers with dedup keys prevent double-charging or corrupt state, and dead-letter queues park poison messages.

No. Function instances are frozen and destroyed outside your control, and local state is never shared across concurrent instances. Persist all durable state in a managed database or cache and treat every invocation as fresh.