Serverless Horrors

Original Article ↗ Hacker News Discussion ↗

Surprise Bills and Recourse

  • Many anecdotes of 4–6 figure surprise bills (AWS, GCP, Azure, Oracle, Vercel, Firebase, Netlify, etc.), often from test or hobby projects unintentionally exposed to high traffic or misconfigurations.
  • Several posters say large clouds usually waive or heavily reduce such bills if you open a support ticket, but there’s no clear, published guarantee; fear remains of being the edge case that gets pursued for payment.
  • Some note social‑media shaming as the only reliably fast escalation path; others report successful quiet resolutions via support, especially for enterprise customers.

Lack of Hard Spend Limits

  • Repeated criticism: major providers offer only budget alerts, not real, synchronous hard caps; billing data often lags by hours or days.
  • People want “cut me off at $X and freeze services” as a first‑class, obvious setting, especially for free tiers and small projects.
  • Counter‑argument: implementing real‑time caps at scale is technically complex and risks data loss or unintended downtime; provider incentives likely also discourage it.

Security, Misconfiguration, and “Denial of Wallet”

  • Many stories are rooted in open S3 buckets, direct origin access bypassing CDNs, missing rate limits, recursive or runaway serverless calls, verbose logging, or insecure defaults in third‑party tools.
  • Some argue this is primarily user error and poor architecture; others reply that tools which allow a 10,000× cost escalation without guardrails are inherently dangerous.

Serverless Development Pain

  • Several engineers describe large Lambda/Cloud Functions backends as hard to debug and test locally, with black‑box behavior, cold starts, and environment mismatches.
  • Workarounds include per‑developer stacks, local emulators, tools like LocalStack/SST, but iteration is still slower than with a traditional app on a VM or container.

VPS / Bare Metal vs Cloud Economics

  • Strong contingent prefers fixed‑price VPS or bare‑metal (Hetzner, DO, etc.) for personal projects and early startups: predictable cost, natural hardware limits, and “failure via downtime, not bankruptcy.”
  • Others note clouds help real businesses survive peak traffic and marketing spikes that would overwhelm a cheap VPS; trade‑off is cost and complexity.

Terminology, Marketing, and Ethics

  • Debate over “serverless” as a misleading or even “Orwellian” term vs a reasonable shorthand for “no server management.”
  • Some see pricing and lack of caps as dark patterns optimized for over‑spend; others frame it as powerful but dangerous tooling requiring competence and responsibility.
  • Ideas raised: regulatory caps on pay‑per‑use services, insurance for runaway cloud bills, and more honest onboarding that emphasizes financial risk.