Fastly runs one of the most technically respected CDN and edge compute platforms — Compute@Edge ships WebAssembly workloads to 70+ POPs globally with millisecond cold starts. The interview is networking-heavy, low-level, and rewards engineers who genuinely care about latency budgets in the single-digit microseconds.
Process
Recruiter screen → 60-minute technical phone (DSA + systems trivia) → onsite virtual: 2 coding (one DSA, one systems-flavored), 1 system design (always edge or networking), 1 deep-dive on a past project, 1 behavioral. Senior+ candidates get an additional architecture round. Cycle: 3–4 weeks.
What they actually ask
- Design a global CDN with origin shielding, request collapsing, and cache invalidation. Focus on the cache hierarchy — edge POPs backed by a shield tier that absorbs origin traffic — and trace what happens on a cache miss. Interviewers push on request collapsing (coalescing simultaneous misses for the same key into one origin fetch) and on invalidating content worldwide by URL or surrogate key without triggering a thundering herd.
- Design Compute@Edge — how do you isolate WASM workloads with sub-millisecond cold start? Compare WebAssembly isolation with containers and V8 isolates, and explain why instantiating a fresh instance per request beats keeping warm processes on cold-start latency. Expect follow-ups on per-request memory limits, stopping one tenant from starving another, and what a compiled module is allowed to touch on the host.
- Design a real-time logging pipeline that handles 1M+ events per second per POP. Walk through batching and back-pressure at the edge, buffering to local disk when the collector is unreachable, and sampling versus full fidelity. The point they probe is keeping the fast path non-blocking so logging never adds latency to the request it records.
- Coding: low-level systems problems — buffer management, state machines, parsing. These reward tight, allocation-aware code over clever algorithms: think ring buffers, an HTTP header parser, or a state machine for a streaming protocol. Practice handling partial input and reusing memory instead of allocating on every call.
- Networking deep dive: TCP, TLS handshake, HTTP/2 vs HTTP/3, BGP basics. Know the round trips cold — a full TLS 1.3 handshake, 0-RTT resumption, and why HTTP/3 over QUIC sidesteps the head-of-line blocking that hurts HTTP/2 on lossy links. BGP usually comes up around anycast: how one IP maps to the nearest POP and what happens when a POP withdraws its route.
Levels and comp (2026)
- SE II: $170K–$210K total
- Senior SE: $250K–$320K
- Staff: $360K–$470K
- Principal: $500K–$650K
Prep priorities
- Read Fastly engineering blog — the posts on Compute@Edge, request lifecycle, and origin shielding are interview gold
- Brush up on Rust (much of the edge runtime is Rust), VCL, and WebAssembly basics
- Networking fundamentals: be able to explain TCP slow start, congestion control, and TLS 1.3
Frequently Asked Questions
Is Fastly more like a networking or a software company?
Both. Software engineering rigor with networking domain depth. You will see questions you would never get at a typical SaaS company.
What languages does Fastly use?
Rust dominates the edge runtime; Go for control plane; Ruby for legacy services; VCL for customer-facing config.
How does Fastly compensation compare to Cloudflare?
Cloudflare pays slightly more on equity post-IPO. Base is comparable. Fastly senior comp lands ~85–90% of equivalent Cloudflare level.
Similar company guides
Prepping for Fastly? Put it to work:
