Object Storage
An S3-shaped store: durable bytes on one path, fast metadata on the other.
Keep puts and gets flowing through a nightly backup window while metadata blips and a gateway node dies.
Every request needs a metadata lookup and an object-bytes operation — the gateway splits them immediately. A cache in front of the metadata store keeps lookups off the database, and a replicated placement fleet in front of the blob store is what survives a backup window's write flood.
Components in play
- Edge LB — Spreads client traffic across the API gateway fleet.
- API gateway — Splits every request into a metadata lookup and an object operation.
- Placement service — Resolves which blob-store shard holds (or will hold) the object.
- Object store — Durable storage for the actual object bytes.
- Metadata cache — Serves hot object metadata from memory so the metadata store stays idle.
- Metadata store — Source of truth for object metadata — keys, sizes, locations.
Graded on these SLOs
- ≥ 96.00%Puts and gets succeed
- ≤ 140 msRequest p99
- ≥ 14k rpsSustain the backup window
- ≤ 90%Fleet headroom
- ≤ 13Lean footprint
The brief
Design an object storage service like S3: clients put and get arbitrary-size objects by key, and every single request also needs a metadata lookup — does this key exist, how big is it, which shard holds it. The two concerns have completely different shapes: metadata lookups are small and latency-sensitive, object bytes are large and durability-sensitive. They should never share a bottleneck.
Functional
- Put an object's bytes under a key
- Get an object's bytes back by key
- Look up an object's metadata (size, location, existence) on every request
Non-functional
- Durability — an accepted object is never lost
- Metadata lookups must stay fast even while object bytes are under heavy write load
- High availability — puts and gets must survive a metadata blip or a lost gateway node
Warm-up: design decisions
Optional theory to prime the calls a senior engineer would make before you build.
Solving is a Pro feature
Unlock the on-canvas AI design coach, per-run design reviews, the chaos simulator and all 12 problems for $15/month.
- AI design coach on your canvas — pins failure points & over-engineering
- Honest AI design review on every run
- All 12 problems + unlimited chaos simulator
- Points, levels & leaderboard