TL;DR
- Segment: 140 microservices → monolith (velocity recovered)
- Prime Video: serverless → monolith (90% cost reduction)
- Instagram: 14M users with 3 engineers on PostgreSQL + Python
- Pattern: start boring, migrate when you have proof
The Known Failure Modes Principle
When Postgres fails, Stack Overflow has the answer. When your custom database fails, you're on your own.
Boring technology—mature, widely-adopted, battle-tested—has a decisive advantage: its failure modes are documented. Thousands of engineers have hit the same problems and shared solutions.
Novel technology offers no such safety net. When things break, you're pioneering the debugging.
Case Study: Segment's Microservices Retreat
Segment adopted microservices early. The architecture grew to 140 distinct services.
What happened: A change to a shared library required redeploying all services. Testing became a nightmare. Local development required spinning up dozens of containers.
The reversal: They consolidated back to a monolith. Simplified testing, recovered engineering velocity, reduced operational complexity.
The lesson: Microservices solve organizational problems, not technical problems. For teams under 50 engineers, a modular monolith is almost always right.
Case Study: Amazon Prime Video
Prime Video's audio/video monitoring service was built on Lambda and Step Functions.
The problem: Data transfer between distributed components became expensive. They hit scalability ceilings.
The pivot: Refactored to a single monolith on ECS.
The result: 90% infrastructure cost reduction.
The Counter-Intuitive Truth
| Company | Initial Stack | Scale Achieved |
|---|---|---|
| PostgreSQL + Python | 14M users with 3 engineers | |
| Shopify | Ruby on Rails | From launch through IPO |
| GitHub | Ruby on Rails | The world's code repository |
These companies chose boring technology because developers were immediately productive, hiring was easy, and they could focus on product instead of infrastructure.
Key Takeaways
- Start with boring, migrate when you have data proving you need to
- Microservices are for organizational scaling, not technical scaling
- The cost of debugging novel technology compounds faster than its benefits
- Instagram scaled to 14M users with PostgreSQL—your startup can too
Originally published on alexmayhew.dev
Top comments (0)