Cloudflare interview process (2026)
Round-by-round breakdown of the Cloudflare software engineering interview across 3 levels and 15 rounds: what each round covers, what interviewers evaluate, sample questions, and preparation tips.
Software Engineer Intern
Difficulty: Medium · Timeline: ~3-4 weeks · Recommended prep: 4-6 weeks
-
Round 0: Application + AI-app fast-track and recruiter chat
Phone · ~30 min
Focus: Resume, AI-powered project on Cloudflare Workers, motivation, logistics
Cloudflare's 1,111-intern program includes Bengaluru SWE intern seats (12-16 week internship, 3-5 days in office). Applications that attach a small AI-powered application built on Cloudflare (Workers, Workers AI, Pages) get fast-tracked review. A recruiter call then covers your degree program, projects, preferred stack (Go, Rust, Python, C/C++, TypeScript are all fine) and internship logistics.
What they evaluate: Genuine interest in how the internet works, a project you can talk about in depth, basic communication. The optional Workers project is the single strongest differentiator at this stage.
Tips: Spend a weekend shipping a tiny Workers AI app (even a simple RAG or URL summarizer) and link it in the application - Cloudflare explicitly fast-tracks these. Deploy it on a workers.dev subdomain so the link just works.
-
Round 1: Online coding assessment (HackerRank)
Assessment · ~60-90 min · difficulty 2/5 · elimination round
Focus: DSA easy-medium, hash maps, sorting, strings, log/text parsing
HackerRank-style assessment with 1-2 problems. Cloudflare leans practical rather than puzzle-y: parsing structured input (log lines, headers), frequency counting, top-K, and standard easy-medium DSA. Any mainstream language is accepted.
What they evaluate: Correct, working code under time pressure; comfort turning messy input into clean data structures; edge-case handling (empty input, ties, malformed lines).
Sample question: You are given N web-server access log lines of the form "<ip> <method> <path> <status>" and an integer K. Print the top K IP addresses by request count, one per line as "<ip> <count>", ordered by count descending, breaking ties lexicographically by IP. Input: first line N and K, then N log lines.
Tips: Practice HackerRank input parsing (getline vs cin mixing trips people up). Cloudflare's screens reward clean working code over clever tricks - get a brute-force version passing, then optimize.
-
Round 2: Technical interview: coding + CS fundamentals
Onsite · ~60 min · difficulty 3/5
Focus: Live coding, data structures, networking basics (DNS, TCP/UDP, HTTP), OS basics
Video round with a Bengaluru engineer. One live coding problem in your language of choice, plus CS-fundamentals discussion with a networking flavor: what happens when you type a URL, DNS resolution, TCP vs UDP, what a CDN/proxy does. Interviewers probe how your project from the application actually works.
What they evaluate: Working code with reasonable complexity analysis; genuine understanding of networking/OS basics rather than memorized definitions; curiosity about Cloudflare's edge network.
Sample question: Implement a function that takes a list of (domain, IP) pairs and answers lookups with wildcard support: a query "a.b.example.com" should fall back to "*.example.com" if no exact entry exists. Discuss complexity and how real DNS resolution differs.
Tips: Revise the classic "what happens when you type cloudflare.com in a browser" end-to-end - DNS, TCP handshake, TLS, HTTP caching. Reading 2-3 Cloudflare engineering-blog posts (Anycast, 1.1.1.1) gives you vocabulary interviewers respond well to.
-
Round 3: Hiring manager + behavioral chat
Behavioral · ~30 min · difficulty 1/5
Focus: Projects walkthrough, teamwork, why Cloudflare, internship expectations
Conversational close-out with the intern host manager. Walks through your resume projects, how you handle feedback and ambiguity, and what you want from the 12-16 weeks. Interns present their project to the company at the end, so managers look for people who can communicate.
What they evaluate: Clear articulation of what YOU built in team projects, coachability, and authentic interest in internet infrastructure over brand-name collecting.
Tips: Have one 2-minute and one 5-minute version of your best project story. Mention a specific Cloudflare product (Workers, R2, 1.1.1.1) you have actually used when answering "why Cloudflare".
Software Engineer
Difficulty: Medium · Timeline: ~3-5 weeks · Recommended prep: 6-8 weeks · Cooldown: ~6 months
-
Round 0: Recruiter screen
Phone · ~30 min
Focus: Background, motivation, team fit, logistics, compensation expectations
Standard 30-minute call covering your background, why Cloudflare, and which Bengaluru team the req maps to (the India engineering office is newer and growing, with systems- and network-heavy charters). India candidates have reported patchy recruiter responsiveness, so drive scheduling proactively.
What they evaluate: Communication, alignment with the specific role, realistic notice-period and comp expectations.
Tips: Ask which team and stack (Go vs Rust vs C++) the loop is for - Cloudflare loops vary noticeably by team, and knowing this focuses your prep. Follow up politely if you hear nothing for a week.
-
Round 1: Hiring manager screen
Phone · 30-45 min · difficulty 2/5 · elimination round
Focus: Resume deep dive, past projects, team interest, networking fundamentals
Cloudflare runs the hiring-manager conversation earlier than most companies. Mostly resume-driven and friendly, but Bengaluru candidates have been asked pointed technical questions here too - one report includes deep DNS questions and encryption-key handling in the HM round, with an immediate reject after. Treat it as technical, not a chat.
What they evaluate: Depth on your own projects (they drill into architecture decisions you personally made), fundamentals relevant to the team (DNS, TLS, HTTP for edge teams), and motivation for Cloudflare specifically.
Sample question: Walk me through a system you built end-to-end. Follow-ups reported in India loops: how does DNS resolution work from stub resolver to authoritative server, and where do encryption keys live in a TLS termination path?
Tips: Re-read your resume as an adversary: every line is fair game for a "why" three levels deep. Brush up DNS and TLS basics even for a generalist role - this is Cloudflare's home turf.
-
Round 2: Technical screen: practical systems coding
Phone · ~60 min · difficulty 3/5 · elimination round
Focus: Practical coding (rate limiter, LRU variants, ring buffer, parsing), language of choice
Live coding on CoderPad/HackerRank. Candidates consistently describe it as "more HackerRank-practical than LeetCode": implement a rate limiter, an LRU-cache-style modification, a ring buffer, or input-parsing backend feature work. Go, Rust, Python, C++, or TypeScript are all accepted; systems teams like seeing Go/Rust/C++ fluency.
What they evaluate: Choosing the right data structure fast, writing production-shaped code (naming, edge cases, tests you would add), and staying calm - the exact prompt matters less than clean execution.
Sample question: Implement a per-client sliding-window rate limiter: given W (window seconds) and L (max requests), process N requests as (client_id, timestamp) in non-decreasing time order and print ALLOW or DENY for each. Only allowed requests count toward the window. Input: first line N W L, then N lines "client ts".
Tips: Practice implementing rate limiters (sliding window AND token bucket), LRU cache, and a ring buffer from scratch in under 25 minutes each - these exact patterns recur across Cloudflare reports. Narrate trade-offs (memory per client, what changes under concurrency) as you code.
-
Round 3: Onsite coding: build + debug (AI-assisted for some teams)
Onsite · ~60 min · difficulty 4/5
Focus: Deeper practical coding, debugging an unfamiliar codebase, AI-assisted coding
Virtual-onsite coding round(s), 3-5 back-to-back for the full loop. One round extends a component with real-world constraints (e.g., LRU cache with TTL - exactly what an edge cache does). Several 2025-26 reports also describe a debugging exercise on an existing system and an unusual AI-assisted round where you use prompts to find and fix bugs or add a feature in an unfamiliar repo.
What they evaluate: Reading and reasoning about code you did not write, incremental verification while debugging, sensible use of AI tools without losing rigor, and follow-ups on concurrency and memory limits.
Sample question: Extend an LRU cache with per-entry TTL (edge-cache semantics): capacity C, operations PUT key value ttl ts and GET key ts. GET returns -1 if the key is absent or expired (lazily evict expired entries); both PUT and GET refresh recency. Print each GET result. Input: first line C Q, then Q operation lines.
Tips: For the AI-assisted round, practice pairing with an AI tool on an unfamiliar open-source repo: reproduce a bug, form a hypothesis, then verify the AI's fix yourself - they reject blind prompt-accepting. Know classic LRU follow-ups: thread safety (lock striping) and memory bounds.
-
Round 4: System design: edge-scale, no managed cloud
System_design · ~60 min · difficulty 4/5
Focus: Distributed rate limiting, log pipelines, caching, designing without AWS/GCP managed services
Design discussion pitched at mid-level: a distributed rate limiter inside an HTTP proxy, a log-collection pipeline spanning global edge PoPs, or a caching layer. Cloudflare's twist: you are usually asked to design WITHOUT managed cloud services (no S3, no Pub/Sub) - they build their own infrastructure, so expect to reason about Kafka, ClickHouse, disk layouts, and raw servers.
What they evaluate: Requirements clarification, explicit trade-offs (latency vs consistency, cache freshness vs hit ratio), realistic capacity math at "billions of requests/day" scale, and graceful degradation under DDoS-sized load spikes.
Sample question: Design the log pipeline for a CDN with 300+ edge PoPs: every HTTP request generates a log line; customers need near-real-time analytics and 30-day retention. You may not use any managed cloud service. Cover transport from PoPs, backpressure when a region is cut off, aggregation, and the storage engine choice.
Tips: Rehearse designs from Cloudflare's own blog: their logging pipeline (Kafka + ClickHouse) and rate-limiting posts map almost 1:1 to reported questions. Lead with the failure story - what happens when a PoP loses backhaul - before being asked.
-
Round 5: Behavioral ("Orange Cloud") round
Behavioral · ~30 min · difficulty 2/5
Focus: Culture fit, curiosity about the internet, ownership, transparency, mistakes
Cloudflare's named culture round. Situational questions: a production mistake you made and how you handled it, a technical disagreement, explaining something complex to a non-technical person, and why Cloudflare / why this team. Sometimes folded into the hiring-manager conversation.
What they evaluate: Bias for action, intellectual curiosity about how the internet works, transparency about failures (they explicitly value owning mistakes - it is part of the company's founding story), and mission alignment with "helping build a better Internet".
Tips: Prepare STAR stories for: a mistake in production, a disagreement you lost gracefully, and a time you dug into something out of pure curiosity. Honest specifics beat polish here.
Senior Software Engineer
Difficulty: Hard · Timeline: ~4-6 weeks · Recommended prep: 8-10 weeks · Cooldown: ~6-12 months
-
Round 0: Recruiter + hiring manager screens
Phone · 2 x 30-45 min · difficulty 1/5 · elimination round
Focus: Track record, scope of past ownership, team charter fit, why Cloudflare now
Two quick conversational rounds, usually within the first week. The HM screen carries real weight at senior level: expect probing on systems you owned, on-call and incident stories, and fluency in the team's domain (edge proxy, DNS, DDoS, storage, or Workers runtime - Bengaluru hosts several systems-heavy charters). Senior loops average ~4 weeks end to end.
What they evaluate: Evidence of owning production systems at scale, architectural judgment in your own history, and whether your depth matches the team's stack (Go/Rust/C++ systems work rather than pure CRUD).
Tips: Bring numbers: QPS, p99s, incident counts, cost savings. Senior HM screens at Cloudflare reward engineers who talk about failure modes of their own systems unprompted.
-
Round 1: Technical screen: systems coding
Phone · ~60 min · difficulty 4/5 · elimination round
Focus: Practical coding with systems constraints, concurrency, memory, Go/Rust/C++ fluency
Same practical format as mid-level but with sharper follow-ups: implement a token-bucket or multi-tier rate limiter, a bounded ring buffer, or an API-design-flavored coding task (one Bengaluru senior report: coding round centered on API design). Follow-ups go to concurrency: making it thread-safe, lock contention, memory per tenant at 10M clients.
What they evaluate: Production instincts - bounded memory, back-pressure, thread-safety - argued concretely in your language of choice. Seniors are expected to discuss what breaks at Cloudflare scale without prompting.
Sample question: Implement a token-bucket rate limiter supporting burst capacity B and refill rate R tokens/sec, then make it safe for concurrent callers. Follow-ups: memory for 10 million client buckets, lazy refill vs a background ticker, and how you would shard state across proxy processes on one edge server.
Tips: Be ready to reason in Go or Rust idioms even if you code in C++ (mutex vs atomic CAS, sharded locks). Rehearse the lazy-refill token bucket - computing tokens from elapsed time on each request - since it removes the need for timers and interviewers like it.
-
Round 2: System design deep dive: global edge infrastructure
System_design · 60-75 min · difficulty 5/5
Focus: Global KV stores, DDoS mitigation, CDN/DNS at scale, build-it-yourself infrastructure
The signature senior round, repeatedly described by candidates as a genuine engineering discussion rather than an interview. Reported prompts: a globally distributed key-value store (Cloudflare KV), a DDoS mitigation system, handling millions of DNS queries per second, or CDN cache-invalidation. The no-managed-services rule is strict at this level - you design the storage, replication, and messaging layers yourself.
What they evaluate: Depth over breadth: consistency model choices for edge replication (why eventual, where CRDTs help), Anycast implications, attack-resilience, and honest trade-off narration. They probe until you hit the edge of your knowledge and watch how you behave there.
Sample question: Design Cloudflare KV: a key-value store readable from 300+ PoPs with p99 read latency under 10 ms worldwide, writes accepted centrally, and eventual global propagation. No managed cloud services. Cover replication topology, cache invalidation vs TTL trade-offs, cold-key reads from a far PoP, and what a network partition between a PoP and core does to reads and writes.
Tips: Study Cloudflare's public architecture posts on Workers KV, Quicksilver (their edge config store), and their DDoS write-ups - reported questions track their blog closely. Saying "here is how Cloudflare actually does it, and here is where I would deviate" is a strong senior move.
-
Round 3: Architecture retrospective: your system, deep-dived
Onsite · ~60 min · difficulty 4/5
Focus: Past-project architecture walkthrough, incident handling, CDN/DNS domain probing
A project-retrospective round: pick a system you built or currently own and walk its architecture end to end while a senior engineer cross-examines. Bengaluru senior loops mix in domain questions - CDN behavior, DNS internals, TLS termination - against YOUR system: how would it behave behind Anycast, where does it break under a 100x traffic spike.
What they evaluate: Whether the depth on your resume is real: data-flow precision, why-not-X alternatives you considered, scars from real incidents, and ability to map your experience onto Cloudflare's edge context.
Sample question: Take the highest-scale system you have owned. Reported follow-up pattern: where exactly does a request enter, what is the p99 budget per hop, what failed in your worst incident, and what would you redesign today - then "how would this design change running in 300 PoPs instead of 3 regions?"
Tips: Choose ONE system and rehearse drawing it in under 3 minutes, then defend every arrow. Pre-compute your answers on capacity, cost, and the worst incident - vagueness on your own system is the most common senior reject reason.
-
Round 4: Behavioral + leadership ("Orange Cloud")
Behavioral · 30-45 min · difficulty 2/5
Focus: Mentorship, influence without authority, mistakes and transparency, mission fit
Senior version of the culture round: leading through a disagreement, mentoring a struggling engineer, a call you got wrong and what it cost, and pushing back on a manager. Scheduling across US/EU/India interviewers can stretch the loop - reports note occasional last-minute reschedules, so keep buffer.
What they evaluate: Leadership signal without a management title, transparency about failures, and curiosity about internet infrastructure as a mission ("helping build a better Internet") rather than a compensation stop.
Tips: Have a genuine answer for "tell me about a decision you got wrong at scale" - Cloudflare's culture treats owning mistakes publicly as a feature. Follow up politely but persistently on scheduling; India candidates report it as the flakiest part of the loop.
Other companies' interview processes
Worked sample solutions (code, design diagrams, and STAR answers) for every round are available in the FAANGPlus app: open interview processes.