Short answer: treat object storage as the durable home for the original avatar and backup bytes, not as the image-resizing engine. Put resizing in a server-side worker or image service, write the derived avatar as a separate private object, and keep the backup path optimized for large-file throughput. For a fintech system with per-tenant archives and selected-snapshot restore, this separation matters more than the storage brand.
The useful design has two paths. The first is a small, latency-sensitive avatar path: validate, resize, and publish a private derivative through a controlled read URL. The second is a large-file backup path: stream tenant data into immutable application-managed snapshots, then restore one selected snapshot without mixing transformation work into the transfer.…