Dropbox · Primly Community

Dropbox senior / L5 system design interview, what to expect in 2026

sre_sol · 4 replies

Just finished the Dropbox loop last month, leveling for L5. Here's what the system design round actually looked like.

The prompt was open-ended: design a file sync system. Not exactly a curveball given what Dropbox does, but the depth they want is real. I went maybe 10 minutes on requirements/scoping before touching anything architectural, which felt right. The interviewer pushed back when I tried to over-engineer the metadata store too early.

Things they actually probed: How do you handle concurrent edits / conflict resolution. They wanted specifics, not vague CRDTs. Block-level deduplication vs file-level. Tradeoffs around storage cost. How the sync client decides what changed locally (inotify, FSEvents, polling fallback). Scalability for millions of concurrent clients, not just theoretical millions of files.

What I noticed: the interviewer seemed to care more about how I reasoned through tradeoffs than whether I landed on a specific architecture. When I said "I'd use S3 for object storage" they immediately asked what I'd do if S3 went down and whether consistency or availability mattered more for this use case.

The round is 45-50 minutes. Last 5-10 is for your questions, but they'll use that time too if you're mid-discussion.

For prep: distributed file storage, metadata services, vector clocks or operational transforms for conflict resolution, and client-side sync protocols. If you've touched anything Kafka or change-data-capture adjacent that's relevant because they care a lot about event streaming internally.

Leveling-wise, for L5 they want you driving the conversation. If you wait to be told what to explore next, that's probably going to read junior.

4 replies

infra_ines

The FSEvents / inotify question is a good one to be ready for. I got a variant of it at a different storage company, they care about whether you've thought about edge cases like network drives or cross-platform sync. Did they go deep on how the client batches events before sending deltas to the backend?

remote_swe_42

Yeah briefly. They asked whether I'd batch on a time window or event count threshold and what happens if the client crashes mid-batch. Basically stress-testing fault tolerance at every layer.

jp_newgrad

Did they give you a whiteboard or is it all in a doc/shared editor? Asking because I have a Dropbox L4 loop coming up and I'm not sure what format to expect.

remote_swe_42

Google doc shared editor. No formal whiteboard. You can paste rough diagrams in plain text or just describe them. The interviewer was fine with ASCII-level diagrams.