Cloudflare / Concept reference
Which piece owns
which job?
A compact map to keep beside the course.
This reference introduces vocabulary; it does not mark these concepts as mastered. Each definition points back to the lesson that explains it.
Choose by responsibility
| Need | Course component | Reason |
|---|---|---|
| Queryable application records | D1 + Drizzle | Relational queries and account-filtered results. |
| Fast repeated configuration reads | KV | Cache by link ID; accept and manage staleness. |
| Large evidence files | R2 | Store artifacts separately and connect them by evaluation ID. |
| Independent event processing | Queues | Decouple delivery from consumption and retry failed work. |
| Resumable sequence of operations | Workflows | Track named steps and persist intermediate progress. |
| State for one named coordinator | Durable Objects | Associate scheduling or viewer coordination with an identity. |
| Updates pushed to viewers | WebSockets + tracker object | Broadcast updates to connected clients. |
| Identity and access checks | Better Auth + server middleware | Resolve sessions and scope access using verified identity. |
| Payment lifecycle | Stripe + webhook integration | Synchronize subscription events; add entitlement enforcement. |
Trace a failure to the right boundary
- The page loads but data is fake: inspect the tRPC procedure and shared query; the starter begins with placeholders.
- A binding exists in TypeScript but fails at runtime: inspect the selected environment, actual binding configuration, and entrypoint initialization.
- A changed link redirects to an old URL: inspect cached configuration and invalidation, then the authoritative database record.
- The redirect works but analytics are missing: inspect the producer send, consumer validation, errors, retries, and dead-letter path.
- The AI reports unknown availability: inspect the saved page text and screenshot before rewriting the prompt.
- A workflow fails after screenshot support: inspect non-stream step-result size and apply lesson 34’s R2-first artifact handling.
- A new live-map tab is empty: distinguish the tutorial’s initial-history gap from failure to receive new events.
- Checkout succeeds but the plan does not update: inspect the matching Stripe mode, webhook delivery, signing secret, and subscription record.
- A paid feature is available to everyone: inspect server entitlement checks; billing alone does not enforce a quota.
Added diagnostic prompts based on the course demonstrations; use the relevant lesson to verify the actual failure in your implementation.
Vocabulary
- Worker
- Server-side code invoked by requests or other events; the hosting unit for the course’s two services. Lesson 04
- Entrypoint
- The exported handler or class through which the runtime invokes an application, workflow, or object. Lesson 05
- Binding
- A configured connection exposed to Worker code for accessing a resource or another service. Lesson 06
- Monorepo
- One repository containing multiple applications and shared packages; shared source does not imply one deployment. Lesson 02
- TanStack Router
- The UI’s navigation and route-loading layer. Lesson 07
- TanStack Query
- Client-side management of server data, including cached queries and invalidation after mutations. Lesson 07
- tRPC
- Typed communication between the TypeScript frontend and server procedures. Lesson 07
- Zod
- Runtime schema validation used to check actual values at API, queue, and AI-output boundaries. Lesson 23
- D1
- The SQL database for links, click records, evaluations, and authentication/subscription tables. Lesson 08
- Drizzle
- The typed SQL/data-access layer used to query D1 from shared Data Ops code. Lesson 09
- KV
- A key-value store used here as a read cache for link configuration; it is eventually consistent. Lesson 17
- R2
- Object storage used here for rendered HTML, extracted text, and screenshots. Lesson 31
- Queue
- A durable handoff between a producer and consumer once a message has been accepted. Lesson 19
- Dead-letter queue
- A recovery destination for messages that exhaust the configured delivery retries. Lesson 24
- Idempotency
- Designing an operation so repeating the same logical event does not add an unintended extra effect. Lesson 24
- Workflow
- An execution composed of named steps with persisted progress and retry behavior. Lesson 25
- Browser Rendering
- The course’s name for the remote browser service used to load pages and collect evidence. Lesson 27
- Workers AI
- The model service used to classify extracted destination-page content. Lesson 28
- Durable Object
- An addressable stateful coordinator; the same object identity selects the same logical state. Lesson 35
- Alarm
- A persisted future invocation associated with a Durable Object. Repeating work needs an explicit scheduling policy. Lesson 38
- WebSocket
- A persistent two-way connection used here to push click updates to dashboard viewers. Lesson 41
- Service binding
- An internal connection between Workers; it is distinct from a browser’s public URL. Lesson 44
- Authentication
- Establishing the identity represented by a request/session. Lesson 49
- Authorization
- Deciding whether that identity may access a record or perform an operation. Lesson 52
- Webhook
- An HTTP event notification sent by a service such as Stripe to the application. Lesson 56
- Entitlement
- The feature or quota a plan grants; the server must enforce it separately from taking payment. Lesson 59
- Migration
- A controlled change to a database schema or runtime resource definition; the two uses must be distinguished. Lesson 45
- CPU time / wall time
- Active computation versus total elapsed duration, which can include waiting for I/O. Lesson 04
Six quick checks
Choose from memory, then explain why the other two choices do not fit.
1. Where should queryable link records live?
Read the reasoning
D1 holds the relational records. KV caches configuration, while R2 holds larger artifacts. Revisit lesson 08
2. Which component manages a resumable evaluation sequence?
Read the reasoning
Workflows manages named steps and persisted progress. A queue hands events to consumers. Revisit lesson 25
3. What belongs at the server access boundary?
Read the reasoning
The server checks the session and applies account authorization independently of the interface. Revisit lesson 52
4. What should incomplete page evidence lead you to do?
Read the reasoning
Keep an unknown result and inspect stored evidence; a successful HTTP response is not proof of useful product content. Revisit lesson 29
5. Which metadata selects a country-specific destination?
Read the reasoning
The visitor’s inferred country drives this routing policy. The serving data center is a different location. Revisit lesson 14
6. What ultimately enforces paid-feature access?
Read the reasoning
Use subscription state in server-side entitlement checks. The course leaves those limits as further work. Revisit lesson 59
Build in layers
Suggested practice sequence: first make one link persist and redirect correctly. Then add one observable queued event. Next add an evaluation whose saved evidence you can inspect. Add coordinated scheduling and live updates only after those paths make sense. Finish with explicit identity, billing synchronization, entitlement rules, and tests for the failures you can name. This is added study guidance, not a claim that all those final checks are completed by the recording.