Cloudflare Coding Interview: Questions & Format

Cloudflare runs a global network of data centers, the edge, that sits between visitors and the sites they load. It handles a large slice of the world's web traffic. The coding loop is built to find people who can extend that kind of infrastructure. So the problems look less like abstract puzzles and more like the components Cloudflare operates: rate limiters, caches that expire, log parsers, request schedulers. The environment is a shared coding pad, usually CoderPad or a HackerRank-style workspace. Your code runs while you write it.

Underneath the framing the substrate is ordinary medium algorithm work: arrays and strings, hash-map design, intervals, heaps, and a little graph and dynamic programming. What changes is how a round unfolds. You often do not get one fixed problem to solve and walk away from. You build a working component, then grow it in layers as the interviewer raises the bar. Cloudflare says little publicly about its loop, and fewer candidates write it up than at a big-tech giant. So treat the counts here as directional rather than precise.

The Cloudflare angle

The single most on-brand problem is a rate limiter you build live and then keep growing. Cloudflare sells rate limiting as a product, so it interviews on it. A recurring version hands you a function shaped like isRateLimited(ip, method, path, userId, tenantId) and asks for a global request counter first. Once that works, scope it per path, then per user and path, then per user, path, and tenant. The signature never changes; the key you count against gets richer each step. The build itself is a hash map and a time window. The interview is everything layered on top: which windowing strategy you pick, how the key space stays bounded, and what happens when two requests race. Drill this one until growing it from a global counter to a composite key is reflex.

How the coding rounds run

The loop opens with recruiter and hiring-manager conversations, then reaches the coding rounds in two shapes. A technical phone screen comes first: a single live problem in the language you are cleanest in, with a small extension bolted on rather than a second full problem. The onsite adds a longer pair-programming round, where the build-then-extend mechanic really lives. Senior loops especially add a separate debugging and code-review round, where you read someone else's code before writing any of your own. Counts and ordering flex by team and level, so read the shape below as a likely pattern.

• Platform: CoderPad or HackerRank. The phone screen and onsite backend rounds run in a shared pad where code executes. Some pipelines add a short online assessment of two or three questions; many skip it and go straight to a live screen. Front-end candidates work in their own local environment on a provided repo.
• Phone screen: ~45-60 min, one problem. One medium algorithm problem or one practical task such as a rate limiter, in the language you choose, plus a small follow-up. A recurring backend warm-up starts at the triplet family and extends it a dimension further.
• Onsite pairing: 60-90 min, build then extend. You implement a working component and grow it as the interviewer raises the bar: scale it, make it thread-safe, handle disconnects. Clean, testable code and the scaling conversation are graded together, and coding blurs into low-level design.
• Debugging / code-review: Senior loops, ~90 min. Review a snippet, enumerate gaps and bugs, then reason about debugging a latency spike across a distributed system using logs and metrics. More common at senior and above; varies by team.

The Cloudflare angle

Cloudflare's stance on AI assistants is split by round, so confirm the rule for each one with your recruiter. Some 2025 and 2026 backend and debugging loops lean into it. Candidates describe a round that turns entirely into how you use coding agents day to day: what tools you reach for, and where they fall short. A separate variant has you inspect an unfamiliar codebase through prompts, then fix a bug. The front-end app-coding round goes the other way and bans AI tools outright, while still allowing search and documentation. So the prep splits. Come ready to talk fluently about directing and critiquing an agent, and ready to hand-write React and TypeScript fast with nothing helping you.

What the problems favor

No standardized question set carries across teams, so what recurs is a posture more than a list. Group the recurring problems by the muscle they work and a rough shape appears. Hash-map design and caching lead. So much of what Cloudflare runs at the edge is a keyed counter or an expiring cache, and that gravity pulls the loop toward keyed data-structure design more than any other family. Array, string, and hashing work fills the phone screen and the log-parsing builds. Intervals, scheduling, graphs, and light dynamic programming round out the rest. These weights are not published anywhere, so read the chart as relative prominence, not a rate any single loop will match.

The Cloudflare angle

A working solution with no scaling story can still fail the round. Cloudflare runs its own metal across hundreds of points of presence, the physical sites where its network meets the public Internet, so the follow-up is rarely academic. After your component works, expect the room to ask how it survives concurrent access, bounded memory, a worker dropping mid-job, or stale metadata. A distinctive twist is the no-managed-cloud constraint. Reach for a hosted queue or object store, and you may be told to redo it on your own infrastructure, building the storage and messaging yourself. One candidate solved the core problem cleanly and still lost the round on the scale conversation. So rehearse the growth. Do not just make it pass.

Two graded stances run in the same loop, and mixing them up costs you. Backend pairing rewards craft over speed: clean, testable, extensible code and a real conversation about scale beat a fast but brittle answer. The front-end app-coding round inverts that emphasis. It hands fullstack and front-end candidates a React and TypeScript repo with mock UI and API, plus about four tasks under the clock. Here finishing counts as much as correctness, and completing two of four has been reported as a reject. On language, most backend rounds are yours to choose, though Go and Rust carry real weight for infrastructure roles given the systems heritage of the codebase. Front-end pipelines are JavaScript and TypeScript.

How these lessons are organized

Each lesson below takes one part of that surface and works its recurring problems as runnable builds, with the Cloudflare framing and the follow-ups the room piles on. Where a lesson rests on a base data-structures idea, the deep walkthrough lives in the free Foundations course so these pages can stay on what is specifically Cloudflare. Start with arrays and strings, the widest phone-screen bucket, then move into the hash-map and caching lesson where the rate limiter lives.

Common questions