Design a File Storage System

A file record needs owner, logical name, object key, size, content type, checksum, lifecycle state, retention policy, and access policy. The object store holds byte blocks or objects; the metadata store answers listing, permission, version, and lifecycle questions. Keeping them separate avoids putting huge payload semantics into ordinary transactional queries.

Define whether files are immutable, versioned, soft-deleted, or immediately erased. Those rules affect caching, legal retention, user recovery, and the possibility of reusing an object key.

For large files, the metadata service can issue a short-lived upload authorization scoped to a destination, size, content type, and expiry. The client uploads parts directly to storage, then finalizes. This keeps application servers from becoming a bandwidth bottleneck while retaining control over what object may be written.

Multipart upload needs a resumable session, ordered or indexed parts, checksums, an expiry cleanup path, and a finalization rule that marks a file available only after the object is complete and verified.

Ask how many copies, failure domains, and regions are needed before calling an upload durable. The answer affects cost, write acknowledgment latency, and recovery time. A backup is not enough if its restore path has never been measured or if metadata and bytes cannot be reconciled after a partial failure.

Lifecycle rules can move infrequently accessed objects to cheaper storage, but retrieval latency, deletion semantics, and legal holds must remain visible to the product. Avoid treating archival as a transparent implementation detail when users may need data quickly.

At staff scope, distinguish control-plane availability from data-plane availability. Users may be unable to list files while a direct object download still works, or vice versa. Define security boundaries, encryption key ownership, audit trails, recovery drills, and how a regional event affects each plane.

In an interview, evolve from a single object store and metadata table to multipart upload, background processing, replication, lifecycle automation, and regional replication only when the workload and recovery objectives require them.