Machine Coding Question: Design an S3-Like Object Storage System

S3 doesn't have real folders or subfolders. Something that looks like this:

photos/
  └── 2024/
      ├── beach.jpg
      └── mountain.jpg
docs/
  └── report.pdf

is actually just three objects, each with a key — photos/2024/beach.jpg, photos/2024/mountain.jpg, docs/report.pdf. The / is part of the key string with no structural meaning; the S3 console and SDKs split keys on / to simulate folders, but behind the scenes S3 is a flat, massive, distributed key-value store.

This one tests whether you can design a clean key-value storage abstraction, layer ownership and access control on top of it, and keep the whole thing thread-safe for concurrent users — without ever needing folders as a first-class concept.

Requirements are split into three tiers so you know what to prioritize under time pressure: build first, build if time allows, and discuss only.

Everything must be stored in memory (no external DB). The design must be thread-safe to support concurrent users, built with clean, modular, object-oriented design, and left open to adding further AWS-style services in the future.

Extensibility checks — can your design layer access control and history on top of buckets and objects without a rewrite.

Alice registers, logs in, and manages a bucket end to end — creating it, uploading and listing objects by key, downloading one, and deleting objects before the (now-empty) bucket itself is removed. The attempt to delete a non-empty bucket is correctly rejected.