[Miniflare 3] Re-implement R2 gateway and add support for multipart uploads using new storage system

[mrbbot]

This PR builds on #555, and re-implements the R2 gateway using the new storage system. This was the last gateway we needed to migrate. 🙂 It also adds support for multipart uploads, similar to #486, but now with transactions provided by SQLite and the streaming support of the new blob store.

This is split into two commits, one for re-implementing what we had already with the new storage system, and one for extending that with multipart support. You may want to review the SQLite schema before everything else:

miniflare/packages/tre/src/plugins/r2/schemas.ts

Lines 4 to 71 in d6e2989

	export interface ObjectRow {
	key: string;
	blob_id: string | null; // null if multipart
	version: string;
	size: number; // total size of object (all parts) in bytes
	etag: string; // hex MD5 hash if not multipart
	uploaded: number; // milliseconds since unix epoch
	checksums: string; // JSON-serialised `R2StringChecksums` (workers-types)
	http_metadata: string; // JSON-serialised `R2HTTPMetadata` (workers-types)
	custom_metadata: string; // JSON-serialised user-defined metadata
	}
	export const MultipartUploadState = {
	IN_PROGRESS: 0,
	COMPLETED: 1,
	ABORTED: 2,
	} as const;
	export interface MultipartUploadRow {
	upload_id: string;
	key: string;
	http_metadata: string; // JSON-serialised `R2HTTPMetadata` (workers-types)
	custom_metadata: string; // JSON-serialised user-defined metadata
	state: ValueOf<typeof MultipartUploadState>;
	// NOTE: we need to keep completed/aborted uploads around for referential
	// integrity, and because error messages are different when attempting to
	// upload parts to them
	}
	export interface MultipartPartRow {
	upload_id: string;
	part_number: number;
	blob_id: string;
	size: number; // NOTE: used to identify which parts to read for range requests
	etag: string; // NOTE: multipart part ETag's are not MD5 checksums
	checksum_md5: string; // NOTE: used in construction of final object's ETag
	object_key: string | null; // null if in-progress upload
	}
	export const SQL_SCHEMA = `
	CREATE TABLE IF NOT EXISTS _mf_objects (
	key TEXT PRIMARY KEY,
	blob_id TEXT,
	version TEXT NOT NULL,
	size INTEGER NOT NULL,
	etag TEXT NOT NULL,
	uploaded INTEGER NOT NULL,
	checksums TEXT NOT NULL,
	http_metadata TEXT NOT NULL,
	custom_metadata TEXT NOT NULL
	);
	CREATE TABLE IF NOT EXISTS _mf_multipart_uploads (
	upload_id TEXT PRIMARY KEY,
	key TEXT NOT NULL,
	http_metadata TEXT NOT NULL,
	custom_metadata TEXT NOT NULL,
	state TINYINT DEFAULT 0 NOT NULL
	);
	CREATE TABLE IF NOT EXISTS _mf_multipart_parts (
	upload_id TEXT NOT NULL REFERENCES _mf_multipart_uploads(upload_id),
	part_number INTEGER NOT NULL,
	blob_id TEXT NOT NULL,
	size INTEGER NOT NULL,
	etag TEXT NOT NULL,
	checksum_md5 TEXT NOT NULL,
	object_key TEXT REFERENCES _mf_objects(key) DEFERRABLE INITIALLY DEFERRED,
	PRIMARY KEY (upload_id, part_number)
	);
	`;
	// NOTE: `object_key` foreign key constraint is deferred, meaning we can delete
	// the linked object row, *then* the multipart part rows in a transaction,
	// see https://www.sqlite.org/foreignkeys.html#fk_deferred for more details

Closes DEVX-592 and DEVX-593

[changeset-bot]

⚠️ No Changeset found

Latest commit: 52c7c01

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[petebacondarwin]

LGTM with a few comments. A lot of the second commit appeared to be just replacing clock with timers, which would have been nice in a separate commit.
Nice work!

[mrbbot]

Rebasing...