Private Image Thumbnails: Signed URLs, Worker Resizing, and Retention Costs

The article recommends keeping originals and thumbnails in a private object store, resizing them in a Node.js worker using Sharp, and serving results through short-lived signed URLs. It argues that the storage service should act as the origin and durable store rather than as an image transformer or public host, which matters especially in regulated fintech settings. Cost is driven mainly by retained and transferred bytes, so the author advises inventorying storage and egress for each object class before choosing a bucket layout. Deletion policy is identified as the largest cost lever, with the trade-off that early deletion of originals prevents later regeneration or investigation. The piece also describes an idempotent, predictable-key workflow in which existing derivatives are checked before resizing, supported by a database index since prefix-based listing is the only search available.