6.3 The Mental Model to Keep
Think in concepts, map vendors onto them, and know which questions to ask.
Product names change. Concepts do not. The most durable thing to take from this part of the handbook is a mental model that survives changing employers, clouds and frameworks.
Think in concepts, not products
Instead of memorizing:
CloudFront · ALB · ECS · Route53hold the shape:
DNS
↓
CDN
↓
Load Balancer
↓
Compute
↓
StorageThen map vendors onto it as needed:
| Concept | AWS | Cloudflare / Others |
|---|---|---|
| DNS | Route 53 | Cloudflare DNS |
| CDN | CloudFront | Cloudflare |
| Compute | ECS / Lambda | Workers |
| Object storage | S3 | R2 |
| Load balancer | ALB | Cloudflare Load Balancing |
| Logs | CloudWatch | Datadog, Grafana, others |
This is what stops you becoming fluent in one platform's vocabulary and helpless outside it. Someone who knows "CloudFront" knows a product; someone who knows "CDN" can evaluate four of them and pick one.
The final path
Diagram source
flowchart LR
Browser --> HTTP
HTTP --> DNS
DNS --> TLS
TLS --> CDN
CDN --> Nginx
Nginx --> Docker
Docker --> CICD[CI/CD]
CICD --> AWS
AWS --> Observability
Observability --> Terraform
Terraform --> KubernetesWhat this is for
A strong frontend engineer does not need to operate all of this. But looking at a stack like:
Browser
↓
CloudFront
↓
ALB
↓
ECS
↓
API Gateway
↓
Servicesthey should be able to say:
- where requests go
- where caching happens
- where TLS terminates
- where authentication is checked
- where failures can occur
- where logs should appear
- what is frontend responsibility
- what is backend responsibility
- what is infrastructure responsibility
Those last three are the ones that change how a team works. Most arguments about whose bug something is are really arguments about a boundary nobody has drawn.
Questions worth being able to answer
A self-check, drawn from across these chapters:
Delivery. Why is the site slow in Singapore but fast in Virginia? Why does a
user still see old JavaScript after a deploy? Where would you set cache headers
for a hashed bundle, and why immutable?
Routing. A request returns 502. Which layers could have produced it? Why does the app think every request is HTTP when it is served over HTTPS?
Config. Why does it work locally and fail in the container? Which of your environment variables are public, and how would you verify that?
Auth. Why should the session token not be in localStorage? What does
SameSite=Lax prevent, and what does it not?
Operations. Checkout is slow for some users. What do you look at first? Which signal — logs, metrics, traces, errors — answers "how many users are affected"?
If most of those have an answer, the model is working.
Where the line is
You are not expected to run clusters, tune databases or design multi-region failover. Those are real specializations and other people do them full time.
What is expected — and what this part of the handbook was for — is that infrastructure stops being "the mysterious stuff DevOps handles" and becomes part of your normal engineering mental model: a system you can reason about, debug across, and make informed decisions within.
That shift is the whole point. Everything else is details you can look up.