You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Bucket mutation data storage: investigate separate space for block diffs #9
Today, in storacha/sprue#34 every S3 PUT/DELETE produces a CAR carrying the body chunks, the new ObjectManifest, and the mutated MST path nodes. The bucket's "current root" is held in a small SQLite registry (pkg/ms3t/registry/). There is no separate persistent record of "the sequence of bucket mutations" — the MST history is implicit in the chain of CARs in Forge, but there's no separate index over it.
What's worth investigating
A separate space dedicated to bucket mutation block diffs, distinct from the body-data space, opens up: replay, audit, GC, point-in-time reads, billing-by-mutation, federation between ms3t instances.
Per-bucket MST history could be made queryable without walking the whole content space.
Alternatively: continue inlining structural blocks with body data and add a separate index over mutation events instead of a separate storage space.
See open question S3 on Forge #6 in pkg/ms3t/architectural.md ("MST for buckets, registry for buckets") — discussed and deferred during PR #34 because the registry needs SQL-style transactional CAS that Forge doesn't currently provide.
Acceptance
A written proposal as markdown enumerating options, trade-offs (storage cost, retrieval cost, consistency model), and a recommendation
Recommendation linked from the epic and reviewed before any code lands against it
Either the proposal closes this question, or it spawns a follow-up implementation issue
Activity
alanshaw commentedon Jun 24, 2026
Decided to store in bucket.
alanshaw commentedon Jun 24, 2026
Incoming proposals for