21 ways to draw a system

Use when · A process with real yes/no forks, where the branch itself is the point.

S3 upload validation: file arrives → size check → type check → virus scan → accept or reject.

Use when · A process that crosses team or service boundaries and the handoffs matter.

Checkout: Client lane → API Gateway lane → Orders service lane → Payments service lane → Warehouse lane.

Use when · A back-and-forth between a fixed set of participants where ordering and returns matter.

Client → API Gateway → Auth Lambda → DynamoDB → back up the chain with a signed JWT.

Use when · A one-directional process with clear pass/fail gates, no branching back.

CI/CD: Build → Unit tests → Security scan → Staging deploy → Smoke test → Production deploy.

Use when · Something with a defined lifecycle where you need to know what triggers each transition.

EC2 instance: pending → running → (stopping ⇄ stopped) → shutting-down → terminated.

Use when · Explaining how a whole system fits together, for onboarding or a design doc.

Three-tier web app: CloudFront → ALB → EC2/ECS app tier → RDS primary + read replica, S3 for assets.

Use when · Showing scope and containment: what is inside what, not what talks to what.

AWS Region → VPC → Availability Zone → Subnet → Network interface — each one only exists inside its parent.

Use when · Showing a system’s services as independently deployable units, each with a tech label.

Storefront [React SPA] —HTTPS→ API [Node/Express] —gRPC→ Catalog service, —SQL→ Orders DB [Postgres].

Use when · Debugging or documenting exactly how traffic is allowed to move through a network.

VPC with a public subnet (ALB, NAT gateway, IGW) and a private subnet (app servers, RDS) either side of a route table.

Use when · Documenting a schema — what tables exist and how rows in one relate to rows in another.

users (1) —< orders (many) >— order_items >— products — a store schema’s actual foreign keys.

Use when · Helping someone choose between options by walking a series of qualifying questions.

Choosing AWS compute: "Need sub-100ms cold start?" → No → "Stateless & bursty?" → Yes → Lambda.

Use when · Comparing 2–4 options across the same fixed set of criteria.

SQL vs NoSQL across schema flexibility, horizontal scaling, join support, and consistency guarantees.

Use when · Positioning several options against exactly two independent variables at once.

Consistency vs. availability trade-off: DynamoDB, Cassandra, RDS, and etcd placed by where each one leans under partition.

Use when · Showing genuine overlap between sets, where some things belong to more than one.

IAM permissions: what the Deploy role, the ReadOnly role, and the DBA role each grant — and the narrow slice all three share.

Use when · A single either/or call where the reasoning matters more than the visual.

Monolith vs. microservices for a 4-person team: deploy simplicity and low ops cost vs. independent scaling and blast-radius isolation.

Use when · A sequence of dated or ordered events where the gaps between them matter.

Incident postmortem: 03:14 error rate spikes → 03:19 paged → 03:26 root cause found → 03:31 rollback → 03:40 recovered.

Use when · A process where volume drops at every stage and the drop-off rate is the story.

Deploy funnel: 500 commits/week → 410 pass CI → 380 pass review → 350 reach staging → 340 ship to prod.

Use when · Laying out everything that relates to one central idea, with no inherent order between branches.

AWS services radiating from "Compute & storage core": EC2/Lambda/ECS one branch, S3/EBS/EFS another, RDS/DynamoDB a third.

Use when · Reporting a handful of headline metrics where the number itself is the message.

Service health: p99 latency 142ms (good), error rate 0.4% (good), uptime 99.95% (warning — below 99.99% SLO).

Use when · Explaining the internals of a single thing, where position inside it is the information.

A request crossing a load balancer: listener → target group → health check → sticky-session cookie → backend pick, each labelled on a cutaway of the LB.