Blob Store

Git.BlobStore holds everything big: compacted packs, clone bundles, oversize objects, and the spilled bodies of large pushes. The contract is what the pack plane needs and nothing more:

interface BlobStoreShape {
  get(key, range?): Effect<BlobBody | null>; // ranged read
  put(key, body, { contentLength }): Effect<void>; // whole object
  multipart(key): Effect<BlobMultipart>; // streaming write of unknown length
  delete(keys): Effect<void>;
  list(prefix): Stream<BlobMeta>; // purge only, never on a serving path
}

The package owns no bucket. You declare one and pass it in, and the one Layer serves both places bytes are touched: the Worker streaming a clone bundle and the Durable Object writing a pack.

R2

export const GitObjects = Cloudflare.R2.Bucket("GitObjects");

Layer.provide(Git.BlobStoreR2(GitObjects))

The default for a Cloudflare-hosted deployment. No egress fees and no cross-internet hop on the serving path.

S3

export const GitObjects = AWS.S3.Bucket("GitObjects");

Layer.provide(Git.BlobStoreS3(GitObjects))

Compute stays on Cloudflare and bytes live in your S3 bucket. BlobStoreS3(GitObjects) mints a least-privilege IAM identity for the Worker (GetObject, PutObject, DeleteObject, ListBucket, and the multipart actions on that bucket), and signs every request with credentials assumed at runtime. The Layer is built over the S3 bindings, so it runs unchanged in the Worker and in the Durable Object. The stack carries both provider sets:

providers: Layer.mergeAll(Cloudflare.providers(), AWS.providers()),

Every read crosses the public internet, so expect higher serving latency than R2. Reads are coalesced into few large ranges, but the hop is real. Bytes in S3, hashing on Lambda is the full stack.

Your own

Implement the five operations and any object store works. Bring your own store walks through it:

const BlobStoreMine: Layer.Layer<Git.BlobStore> = Layer.effect(
  Git.BlobStore,
  Effect.gen(function* () {
    return {
      get: (key, range) => /* ... */,
      put: (key, body, options) => /* ... */,
      multipart: (key) => /* ... */,
      delete: (keys) => /* ... */,
      list: (prefix) => /* ... */,
    } satisfies Git.BlobStoreShape;
  }),
);

Multipart parts are uniform in size except the last. Both R2 and S3 accept that, and it is what the push path writes.