A Stripe webhook arrives at 3 a.m., three times a month. A VPS runs 24/7 to listen. Cost: fifteen euros fixed plus security patches. A serverless function billed per invocation costs cents — and needs no kernel maintenance. Conversely, a REST API called 500 times per second with a 50 ms p95 requirement hits cold starts and CPU billing that exceeds a small dedicated server.
Serverless is not "no server". It is someone else's server, billed by usage. The advantage appears when load is intermittent, parallelisable and short.
Use cases where serverless wins
Webhooks and integrations — payment, CRM, e-signature: react to events without a permanent daemon.
Media processing — thumbnail, short video transcode, upload antivirus scan: CPU burst then done.
Distributed cron — nightly exports, catalogue sync, log cleanup: cloud scheduler plus function.
Low-traffic API — internal admin, MVP prototype, form to email.
Light edge — header rewrite, A/B test, edge auth (edge computing).
| Use case | Serverless fits | Less fit |
|---|---|---|
| Stripe webhook | Yes | — |
| Main WordPress site | No | Shared VPS |
| High-frequency low-latency API | No | VPS / PaaS |
| Batch of 10k files | Yes (parallel) | Single-thread VPS cron |
| Permanent WebSocket | Hard | Long-lived server |
Limits to anticipate
Cold start — language, package size, VPC attachment. Mitigation: provisioned concurrency (cost), edge Workers.
Timeout — 15 minutes max on many clouds; long jobs → queue plus VPS worker.
Debugging — distributed logs, local replay less trivial than classic dev server.
Vendor patterns — IAM, triggers, proprietary queues. Document export path.
Surprise cost — buggy loop = million invocations. Set quotas and billing alerts.
Realistic hybrid architecture
Common EU project pattern: front and main API on PaaS or VPS; functions for webhooks, images, light ETL; queue between them; managed PostgreSQL with connection pool. You keep control of the business core and delegate decoupled CPU spikes.
The climax: serverless does not remove complexity — it moves it
Decide and move forward without blind spots
First list intermittent versus always-on tasks. Prototype a webhook function with centralised logs. Measure cold start and simulated cost at 10× current traffic. Keep the stateful core on VPS, PaaS or managed DB. Compare EU offers via the directory and PaaS or server.
Frequently asked questions
Does serverless replace my WordPress site?
No for a classic always-on CMS that must serve pages continuously with a permanent PHP core. Yes for side components: image resize on upload, Stripe webhooks or on-demand PDF export — anything that sleeps between events. Keep the main site on shared hosting, VPS or PaaS; serverless complements, it does not replace.
What is cold start?
Latency when waking an idle function after a quiet period. Problematic for an ultra-fast synchronous API; acceptable for async tasks, webhooks or regular traffic that keeps the function warm. Measure it on your runtime and package size — not on a generic demo.
How to handle state and database?
Serverless is stateless by design: state lives in PostgreSQL, DynamoDB or external cache. For DB connections, use a pooler (PgBouncer, RDS Proxy) or short-lived drivers. Without pooling, each invocation may open a connection and exhaust the database under spike load.
Serverless providers in Europe?
AWS Lambda EU regions, Scaleway Functions and Cloudflare Workers cover most cases. Compare jurisdiction, cold start and egress pricing in the directory. Document quotas and billing alerts before production — a buggy loop bills per invocation.
Serverless wins when your load sleeps most of the time — not when your product is a permanent HTTP server.