← home / play
This is a mock e-commerce backend. Click any service to take it down and watch the failure cascade - synchronous calls fail upstream, shared stores fail sideways, queues fail late. Exactly the physics I work on at Meta, minus the pager.
The point of this toy is that the three edge types fail differently - which is why a dependency graph that only shows RPC calls will surprise you in production.
When Auth goes down, the API Gateway that calls it fails next, then everything above it. The failure walks the request path in reverse, fast.
When the Config Store dies, four services degrade at the same instant with no caller-callee ordering. Kill the Cache and watch the read stampede take out the Orders DB.
Kill the Ledger Queue and nothing happens… for a while. The buffer absorbs the hit, then the Ledger Worker starves. Time-to-impact is the tell.
The real version of this idea is in The Service Graph Is a Lower Bound and the fault-injection case study - where the "game" ran against production services and the score was nine figures of protected revenue.