object-storage-patterns

Prefix keys by the dimension you query, scope, and secure on, such as tenant=42/date=2026-07-27/type=invoice/<id>.json. This makes per-tenant listing, per-date lifecycle rules, and per-prefix IAM policies natural, since modern stores auto-scale throughput.

Object storage has no directory nodes, so a rename is a server-side copy to the new key plus a delete of the old one, rewriting all bytes. Renaming a folder of a million objects means a million copy-plus-delete operations.

No, listing is eventually consistent, paginated, and billed per request, so it may miss recently written objects or show deleted ones. Track object existence and metadata in a transactional database and use storage only for bytes.

Choose the class from actual read frequency and latency tolerance, not age alone. Frequently read data in cold tiers incurs retrieval fees exceeding storage savings, and cold tiers bill a minimum storage duration for early deletes.

Use pre-signed URLs: the server mints a short-lived signed URL granting write to one specific object, and the client PUTs directly to storage. Keep buckets private by default and use IAM roles for services instead of embedded credentials.