Sippy: examining a gradual storage migration
Objects copied on demand
On 24 July 2026, Cloudflare extended Sippy to Azure Blob Storage and S3-compatible providers. It copies objects to R2 as the application requests them.
The read path becomes decisive
Our interpretation is that gradual migration must be examined against the project’s access patterns. A media library consulted daily and a rarely opened archive have different profiles. Reading frequency therefore becomes useful information when planning trials.
We would prepare several object categories: frequently read, rarely read, large and missing. For each, check returned content, required metadata and behavior when the source is unavailable. Comparing file counts alone does not prove workloads still function.
Prepare to finish the transition
Also define the criterion for retiring the old storage. We recommend an auditable inventory and checks on important objects, with a documented return plan. Study costs using actual billable reads and operations, without inferring a complete price from a single announcement.
Retrospective watch — July 2026.