A graph models an architecture as programs connected by communication lines: nodes are running services, and edges are SQL queries, HTTP requests, queue messages, or webhooks. section 9.1
The edges often matter more than the nodes because their type predicts failure. Classify each connection as synchronous or asynchronous and internal or public. section 9.1
Use the graph to ask practical architectural questions:
- Name every node and identify who operates its underlying machine. section 9.1
- Trace every wire and predict whether it can time out, fail silently, rate-limit, or return an error. section 9.1
- Mark synchronous edges where users wait and asynchronous edges where work can fail after the response. section 9.1
Checkout architecture
A checkout graph might connect a web server to Postgres, a queue to a worker, and checkout to Stripe. The database edge is internal and synchronous; the Stripe edge is public and synchronous, so vendor latency affects uptime; the queue edge is internal and asynchronous, so failure may be invisible to the user. section 9.1
























