88 / 95 · 05 Performance & Chaos Engineering · Serverless Performance Testing← prev⊞ allnext →☰ Read as one page
13.7When Serverless Is NOT the Right Choice
Performance testing may reveal that serverless is wrong for your use case:
| Signal | Implication |
|---|---|
| Cold starts exceed your latency SLO | Consider containers (ECS, Cloud Run with min-instances) |
| Constant high concurrency (>500 concurrent) | Containers are more cost-effective at sustained load |
| Execution time regularly hits limits (15 min) | Move to containers or batch processing |
| Memory requirements > 10GB | Lambda max is 10GB; use ECS/Fargate |
| GPU required (ML inference) | Use dedicated GPU instances or SageMaker |
The goal of serverless performance testing is not to prove serverless works -- it is to find the boundaries where it stops working and plan accordingly.