roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Prepare for Cloudflare interviews on how a request reaches your origin through the edge and what code runs there. Covers DNS, TLS and caching, edge security, Workers and their data stores.
336 of 336 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 336 new. 0 cut, outside your plan.
Cloudflare rarely gets a round of its own. It comes up inside a platform or SRE conversation about how traffic reaches a service, inside a frontend or full-stack round when the team ships to Workers, and inside a security discussion when the company's origins sit behind the proxy. The opening questions check whether you can follow one request: the visitor's resolver gets a Cloudflare address because the record is proxied, the edge terminates TLS and opens its own connection to the origin, and only then do the cache, the rules and the WAF decide what happens to it. From there the probing works from symptoms. Why does a site loop on redirects after its encryption mode was set to Flexible? Why does an article page report `DYNAMIC` in `cf-cache-status` although its origin sends a long `max-age`? Why are thousands of mobile users blocked together by a rate limit keyed on IP? On the developer side: why does a value just written to KV still read old from another location, and what does routing every request for one ID to a single Durable Object buy a counter? Interviewers listen for the mechanism behind the symptom and for what the fix costs, such as the hit rate a purge throws away or the real visitors a broad challenge rule catches.
The plan follows a site onto the platform before it turns to code. It opens where a move begins, with [delegation](/topics/proto-dns-zones-delegation) and the [record types](/topics/proto-dns-record-types) a zone holds, because [proxy status](/topics/cloud-cloudflare-dns-proxy-status) is a decision made record by record and a nameserver change is what brings a zone onto the network. Encryption comes before caching for a practical reason: until the [encryption mode](/topics/cloud-cloudflare-tls-encryption-modes) matches what the origin serves, requests through the proxy loop or fail, and [termination at a proxy](/topics/proto-tls-termination) is the idea each mode varies. The caching group opens on what any [CDN edge](/topics/found-system-design-cdn-edge) does and on HTTP's own rules. [Cache-Control](/topics/proto-http-caching-cache-control), [Vary and cache keys](/topics/proto-http-caching-vary) and what a shared cache may store are the terms in which Cloudflare's defaults and [Cache Rules](/topics/cloud-cloudflare-cdn-caching-rules) are written, so they come before the settings that override them. Defence then works at the same edge: [the WAF, rate limits and bot controls](/topics/cloud-cloudflare-security) decide what the proxy refuses and how an origin is kept from being reached around it, and [Zero Trust](/topics/cloud-cloudflare-zero-trust) puts identity in front of private applications. The second half is the developer platform. The [function sandbox](/topics/cloud-serverless-execution-model) most serverless platforms use is the baseline [isolates](/topics/cloud-cloudflare-workers-isolates) are measured against, and Pages and the [tooling](/topics/cloud-cloudflare-tooling) sit beside the runtime because Pages Functions run on it and the tooling is how it ships. [Storage](/topics/cloud-cloudflare-storage) and [Durable Objects](/topics/cloud-cloudflare-durable-objects) follow once a Worker needs state that outlives a request, then [queues and Workflows](/topics/cloud-cloudflare-async) once work has to leave the request path. The plan closes on what is built on that runtime: [inference and agents](/topics/cloud-cloudflare-ai), then media delivery and the headless-browser and mail services.
Start from the half the job description points at. When Cloudflare sits in front of an application you run, the first four groups are most of the interview, and a junior candidate is expected to read a record's proxy status and a `cf-cache-status` value without hesitating. When the job is building on Workers, the runtime and its data stores carry the conversation, and the edge groups are worth reading for the request path they put in front of your code. From senior level, both halves turn to ownership: who may change a Cache Rule, how managed WAF rules reach live traffic without blocking customers, whether Tunnel and Access can replace a VPN. Close to a platform or SRE loop, raising importance to must know keeps the questions most Cloudflare conversations reach and drops the product corners; study order can stay on by frequency, where it opens. A Workers role is the opposite case, because its questions go deep into a few branches: keep importance whole and switch study order to by topics, from the runtime down through its data stores. The plan configures zones, rules and Workers through the dashboard, the API and Wrangler; it has no module on keeping a whole Cloudflare account in code through a Terraform provider. How such a setup plans and tracks state is the [Terraform roadmap](/roadmaps/terraform)'s subject.
Cloudflare rarely gets a round of its own. It comes up inside a platform or SRE conversation about how traffic reaches a service, inside a frontend or full-stack round when the team ships to Workers, and inside a security discussion when the company's origins sit behind the proxy. The opening questions check whether you can follow one request: the visitor's resolver gets a Cloudflare address because the record is proxied, the edge terminates TLS and opens its own connection to the origin, and only then do the cache, the rules and the WAF decide what happens to it.
From there the probing works from symptoms. Why does a site loop on redirects after its encryption mode was set to Flexible? Why does an article page report DYNAMIC in cf-cache-status although its origin sends a long max-age? Why are thousands of mobile users blocked together by a rate limit keyed on IP? On the developer side: why does a value just written to KV still read old from another location, and what does routing every request for one ID to a single Durable Object buy a counter? Interviewers listen for the mechanism behind the symptom and for what the fix costs, such as the hit rate a purge throws away or the real visitors a broad challenge rule catches.
The plan follows a site onto the platform before it turns to code. It opens where a move begins, with delegation and the record types a zone holds, because proxy status is a decision made record by record and a nameserver change is what brings a zone onto the network. Encryption comes before caching for a practical reason: until the encryption mode matches what the origin serves, requests through the proxy loop or fail, and termination at a proxy is the idea each mode varies.
The caching group opens on what any CDN edge does and on HTTP's own rules. Cache-Control, Vary and cache keys and what a shared cache may store are the terms in which Cloudflare's defaults and Cache Rules are written, so they come before the settings that override them. Defence then works at the same edge: the WAF, rate limits and bot controls decide what the proxy refuses and how an origin is kept from being reached around it, and Zero Trust puts identity in front of private applications.
The second half is the developer platform. The function sandbox most serverless platforms use is the baseline isolates are measured against, and Pages and the tooling sit beside the runtime because Pages Functions run on it and the tooling is how it ships. Storage and Durable Objects follow once a Worker needs state that outlives a request, then queues and Workflows once work has to leave the request path. The plan closes on what is built on that runtime: inference and agents, then media delivery and the headless-browser and mail services.
Start from the half the job description points at. When Cloudflare sits in front of an application you run, the first four groups are most of the interview, and a junior candidate is expected to read a record's proxy status and a cf-cache-status value without hesitating. When the job is building on Workers, the runtime and its data stores carry the conversation, and the edge groups are worth reading for the request path they put in front of your code. From senior level, both halves turn to ownership: who may change a Cache Rule, how managed WAF rules reach live traffic without blocking customers, whether Tunnel and Access can replace a VPN.
Close to a platform or SRE loop, raising importance to must know keeps the questions most Cloudflare conversations reach and drops the product corners; study order can stay on by frequency, where it opens. A Workers role is the opposite case, because its questions go deep into a few branches: keep importance whole and switch study order to by topics, from the runtime down through its data stores.
The plan configures zones, rules and Workers through the dashboard, the API and Wrangler; it has no module on keeping a whole Cloudflare account in code through a Terraform provider. How such a setup plans and tracks state is the Terraform roadmap's subject.