wire.py
python
sha256:92528ae07d0e1239d87fd5fd1f439e8fbb49c9778a9a400bc4a736073fb28316
feat: byte-range blob reads, file attribution DAG walk, bra…
Sonnet 4.6
minor
⚠ breaking
43 days ago
| 1 | """Wire protocol endpoints — Muse CLI push/fetch transport. |
| 2 | |
| 3 | URL pattern mirrors Git's Smart HTTP protocol: |
| 4 | |
| 5 | muse remote add origin https://musehub.ai/gabriel/muse |
| 6 | |
| 7 | Active endpoints: |
| 8 | |
| 9 | GET /{owner}/{slug}/refs — branch heads + domain metadata (pre-flight) |
| 10 | POST /{owner}/{slug}/push/mpack-presign — get presigned PUT URL for whole mpack |
| 11 | POST /{owner}/{slug}/push/unpack-mpack — server indexes mpack from R2 |
| 12 | POST /{owner}/{slug}/fetch/mpack — server builds and returns fetch mpack |
| 13 | POST /{owner}/{slug}/fetch/presign — presigned GET URLs for large fetches |
| 14 | |
| 15 | These routes MUST be registered before the wildcard UI router in main.py |
| 16 | (/{owner}/{repo_slug}/...) so FastAPI matches the concrete third-segment |
| 17 | paths first. |
| 18 | """ |
| 19 | |
| 20 | import json |
| 21 | import logging |
| 22 | |
| 23 | import msgpack |
| 24 | import pydantic |
| 25 | from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Request, status |
| 26 | from fastapi.responses import Response |
| 27 | from sqlalchemy.ext.asyncio import AsyncSession |
| 28 | |
| 29 | from sqlalchemy import select |
| 30 | from musehub.auth.dependencies import optional_token, require_valid_token, TokenClaims |
| 31 | from musehub.config import get_settings |
| 32 | from musehub.db.database import get_db as get_session |
| 33 | from musehub.db.musehub_abuse_models import MusehubDailyPushBytes |
| 34 | from musehub.db.musehub_collaborator_models import MusehubCollaborator |
| 35 | from musehub.db.musehub_repo_models import MusehubBranch, MusehubObject, MusehubRepo |
| 36 | from musehub.models.wire import WireFetchRequest |
| 37 | from musehub.types.json_types import JSONObject |
| 38 | |
| 39 | from musehub.api.validation import BranchParam, SlugParam |
| 40 | from musehub.rate_limits import limiter, WIRE_PUSH_LIMIT, WIRE_FETCH_LIMIT, OBJECT_LIMIT, REPAIR_LIMIT |
| 41 | from musehub.services.musehub_repository import get_repo_row_by_owner_slug |
| 42 | from musehub.services.musehub_wire import ( |
| 43 | FetchCommitNotFound, |
| 44 | FetchNotIndexedError, |
| 45 | MPackNotReadyError, |
| 46 | FetchNotReady, |
| 47 | MPackValidationError, |
| 48 | ObjectHashMismatch, |
| 49 | record_mpack_bytes_uploaded, |
| 50 | wire_fetch, |
| 51 | wire_fetch_mpack, |
| 52 | wire_fetch_objects, |
| 53 | wire_push_mpack_presign, |
| 54 | wire_push_unpack_mpack, |
| 55 | wire_refs, |
| 56 | wire_repair_commit, |
| 57 | wire_repair_object, |
| 58 | wire_repair_snapshot, |
| 59 | ) |
| 60 | from musehub.storage import get_backend |
| 61 | |
| 62 | logger = logging.getLogger(__name__) |
| 63 | |
| 64 | router = APIRouter(tags=["Wire Protocol"]) |
| 65 | |
| 66 | # ── helpers ──────────────────────────────────────────────────────────────────── |
| 67 | |
| 68 | async def _resolve_repo( |
| 69 | session: AsyncSession, |
| 70 | owner: SlugParam, |
| 71 | slug: SlugParam, |
| 72 | ) -> MusehubRepo: |
| 73 | """Resolve owner/slug → repo row or raise 404.""" |
| 74 | repo = await get_repo_row_by_owner_slug(session, owner, slug) |
| 75 | if repo is None: |
| 76 | raise HTTPException( |
| 77 | status_code=status.HTTP_404_NOT_FOUND, |
| 78 | detail=f"repo '{owner}/{slug}' not found", |
| 79 | ) |
| 80 | return repo |
| 81 | |
| 82 | async def _resolve_repo_id( |
| 83 | session: AsyncSession, |
| 84 | owner: SlugParam, |
| 85 | slug: SlugParam, |
| 86 | ) -> str: |
| 87 | """Resolve owner/slug → repo_id or raise 404 (kept for push — push does its own auth).""" |
| 88 | return (await _resolve_repo(session, owner, slug)).repo_id |
| 89 | |
| 90 | async def _assert_readable( |
| 91 | repo: MusehubRepo, |
| 92 | claims: TokenClaims | None, |
| 93 | session: AsyncSession, |
| 94 | ) -> None: |
| 95 | """Raise 404 if *repo* is private and the caller is not the owner or a collaborator. |
| 96 | |
| 97 | Returns 404 (not 403) to avoid leaking that the repo exists. |
| 98 | """ |
| 99 | if repo.visibility == "public": |
| 100 | return |
| 101 | caller_handle: str | None = claims.handle if claims else None |
| 102 | if caller_handle == repo.owner: |
| 103 | return |
| 104 | # Check collaborators with at least read permission |
| 105 | if caller_handle: |
| 106 | collab_row = (await session.execute( |
| 107 | select(MusehubCollaborator).where( |
| 108 | MusehubCollaborator.repo_id == repo.repo_id, |
| 109 | MusehubCollaborator.identity_handle == caller_handle, |
| 110 | MusehubCollaborator.accepted_at.isnot(None), |
| 111 | ) |
| 112 | )).scalar_one_or_none() |
| 113 | if collab_row is not None: |
| 114 | return |
| 115 | raise HTTPException( |
| 116 | status_code=status.HTTP_404_NOT_FOUND, |
| 117 | detail="repo not found", |
| 118 | ) |
| 119 | |
| 120 | # ── Wire helpers ───────────────────────────────────────────────────────────── |
| 121 | |
| 122 | def _mpack_response(data: JSONObject, request: Request) -> Response: |
| 123 | """Encode *data* as msgpack based on the client's Accept header. |
| 124 | |
| 125 | Clients send ``Accept: application/x-msgpack`` and always receive |
| 126 | binary msgpack. The dict may contain ``bytes`` values (e.g. object |
| 127 | content) which msgpack handles natively. |
| 128 | """ |
| 129 | accept = request.headers.get("accept", "") |
| 130 | if "application/x-msgpack" in accept: |
| 131 | return Response( |
| 132 | content=msgpack.packb(data, use_bin_type=True), |
| 133 | media_type="application/x-msgpack", |
| 134 | ) |
| 135 | return Response(content=json.dumps(data), media_type="application/json") |
| 136 | |
| 137 | def _decode_request_body(raw: bytes, content_type: str) -> JSONObject: |
| 138 | """Decode an HTTP request body from msgpack or JSON. |
| 139 | |
| 140 | Clients send ``Content-Type: application/x-msgpack``; JSON is also |
| 141 | accepted as a fallback for compatibility. |
| 142 | """ |
| 143 | if "application/x-msgpack" in content_type or "application/x-muse-mpack" in content_type: |
| 144 | decoded = msgpack.unpackb(raw, raw=False) |
| 145 | if not isinstance(decoded, dict): |
| 146 | raise ValueError("msgpack body must be a mapping") |
| 147 | return dict(decoded) |
| 148 | parsed = json.loads(raw) |
| 149 | if not isinstance(parsed, dict): |
| 150 | raise ValueError("JSON body must be a mapping") |
| 151 | return dict(parsed) |
| 152 | |
| 153 | # ── wire endpoints ───────────────────────────────────────────────────────────── |
| 154 | |
| 155 | @router.get( |
| 156 | "/{owner}/{slug}/refs", |
| 157 | summary="Get branch heads (muse pull / muse push pre-flight)", |
| 158 | response_description="Repo metadata and current branch heads", |
| 159 | ) |
| 160 | @limiter.limit(WIRE_FETCH_LIMIT) |
| 161 | async def get_refs( |
| 162 | request: Request, |
| 163 | owner: SlugParam, |
| 164 | slug: SlugParam, |
| 165 | _claims: TokenClaims | None = Depends(optional_token), |
| 166 | session: AsyncSession = Depends(get_session), |
| 167 | ) -> Response: |
| 168 | """Return branch heads and domain metadata for a repo. |
| 169 | |
| 170 | Called by ``muse push`` and ``muse pull`` as a pre-flight to determine |
| 171 | what the remote already has. Equivalent to Git's: |
| 172 | ``GET /owner/repo/info/refs?service=git-upload-pack`` |
| 173 | |
| 174 | Private repos are only visible to their owner — unauthenticated callers |
| 175 | receive a 404 (same response as a non-existent repo, to avoid leaking |
| 176 | the existence of private repos). |
| 177 | |
| 178 | Response: |
| 179 | ```json |
| 180 | { |
| 181 | "repo_id": "...", |
| 182 | "domain": "code", |
| 183 | "default_branch": "main", |
| 184 | "branch_heads": {"main": "sha...", "dev": "sha..."} |
| 185 | } |
| 186 | ``` |
| 187 | """ |
| 188 | repo = await _resolve_repo(session, owner, slug) |
| 189 | await _assert_readable(repo, _claims, session) |
| 190 | result = await wire_refs(session, repo.repo_id) |
| 191 | if result is None: |
| 192 | raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="repo not found") |
| 193 | return _mpack_response(result.model_dump(), request) |
| 194 | |
| 195 | @router.post( |
| 196 | "/{owner}/{slug}/push/mpack-presign", |
| 197 | summary="Return one presigned PUT URL for a whole mpack", |
| 198 | status_code=status.HTTP_200_OK, |
| 199 | ) |
| 200 | @limiter.limit(OBJECT_LIMIT) |
| 201 | async def push_mpack_presign( |
| 202 | request: Request, |
| 203 | owner: SlugParam, |
| 204 | slug: SlugParam, |
| 205 | claims: TokenClaims = Depends(require_valid_token), |
| 206 | session: AsyncSession = Depends(get_session), |
| 207 | ) -> Response: |
| 208 | """Return one presigned PUT URL so the client can upload the entire mpack. |
| 209 | |
| 210 | Request body (msgpack): |
| 211 | mpack_key str — sha256:<hex> of the mpack bytes |
| 212 | size_bytes int — advisory byte count |
| 213 | |
| 214 | Response (msgpack): |
| 215 | upload_url str — presigned PUT URL valid for 1 hour |
| 216 | mpack_key str — echoed back for the client to pass to unpack-mpack |
| 217 | """ |
| 218 | raw = await request.body() |
| 219 | ct = request.headers.get("Content-Type", "") |
| 220 | data = _decode_request_body(raw, ct) |
| 221 | mpack_key = str(data.get("mpack_key", "") or "") |
| 222 | size_bytes = int(data.get("size_bytes", 0)) |
| 223 | logger.warning("[mpack-presign] received mpack_key=%s size_bytes=%d content_type=%r", mpack_key, size_bytes, ct) |
| 224 | if not mpack_key: |
| 225 | raise HTTPException( |
| 226 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 227 | detail="mpack-presign requires mpack_key", |
| 228 | ) |
| 229 | # mpack_key must be "sha256:<64-hex>" — anything else will never match |
| 230 | # what the client PUT to MinIO and will fail integrity check in unpack-mpack. |
| 231 | if not mpack_key.startswith("sha256:") or len(mpack_key) != 7 + 64: |
| 232 | raise HTTPException( |
| 233 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 234 | detail="mpack_key must be 'sha256:<64-hex>'", |
| 235 | ) |
| 236 | _settings = get_settings() |
| 237 | if size_bytes > _settings.mpack_max_bytes: |
| 238 | raise HTTPException( |
| 239 | status_code=status.HTTP_413_CONTENT_TOO_LARGE, |
| 240 | detail=( |
| 241 | f"mpack size {size_bytes:,} bytes exceeds limit " |
| 242 | f"{_settings.mpack_max_bytes:,} bytes" |
| 243 | ), |
| 244 | ) |
| 245 | # 4a — per-user daily byte limit |
| 246 | if _settings.mpack_daily_upload_limit_bytes > 0: |
| 247 | from sqlalchemy import select as _select, func as _func |
| 248 | import datetime as _dt |
| 249 | today = _dt.date.today() |
| 250 | _result = await session.execute( |
| 251 | _select(_func.coalesce( |
| 252 | _func.sum(MusehubDailyPushBytes.bytes_uploaded), 0 |
| 253 | )).where( |
| 254 | MusehubDailyPushBytes.identity_id == claims.identity_id, |
| 255 | MusehubDailyPushBytes.date == today, |
| 256 | ) |
| 257 | ) |
| 258 | daily_total = int(_result.scalar() or 0) |
| 259 | if daily_total >= _settings.mpack_daily_upload_limit_bytes: |
| 260 | raise HTTPException( |
| 261 | status_code=status.HTTP_429_TOO_MANY_REQUESTS, |
| 262 | detail=( |
| 263 | f"daily upload limit of {_settings.mpack_daily_upload_limit_bytes:,} bytes reached; " |
| 264 | "try again tomorrow" |
| 265 | ), |
| 266 | ) |
| 267 | await record_mpack_bytes_uploaded(session, claims.identity_id, size_bytes) |
| 268 | await session.commit() |
| 269 | result = await wire_push_mpack_presign(mpack_key, size_bytes) |
| 270 | return _mpack_response(result, request) |
| 271 | |
| 272 | |
| 273 | @router.post( |
| 274 | "/{owner}/{slug}/push/unpack-mpack", |
| 275 | summary="Read an mpack from storage, index all contents into PG", |
| 276 | status_code=status.HTTP_200_OK, |
| 277 | ) |
| 278 | @limiter.limit(WIRE_PUSH_LIMIT) |
| 279 | async def push_unpack_mpack( |
| 280 | request: Request, |
| 281 | owner: SlugParam, |
| 282 | slug: SlugParam, |
| 283 | claims: TokenClaims = Depends(require_valid_token), |
| 284 | session: AsyncSession = Depends(get_session), |
| 285 | ) -> Response: |
| 286 | """Server reads a previously uploaded mpack from MinIO and indexes it. |
| 287 | |
| 288 | Request body (msgpack): |
| 289 | mpack_key str — sha256:<hex> used when the client called mpack-presign |
| 290 | |
| 291 | Response (msgpack): |
| 292 | commits_written int |
| 293 | snapshots_written int |
| 294 | blobs_written int |
| 295 | """ |
| 296 | raw = await request.body() |
| 297 | ct = request.headers.get("Content-Type", "") |
| 298 | data = _decode_request_body(raw, ct) |
| 299 | mpack_key = data.get("mpack_key", "") |
| 300 | logger.warning("[unpack-mpack] received mpack_key=%s raw_body_len=%d content_type=%r", mpack_key, len(raw), ct) |
| 301 | if not mpack_key: |
| 302 | raise HTTPException( |
| 303 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 304 | detail="unpack-pack requires mpack_key", |
| 305 | ) |
| 306 | branch = str(data.get("branch") or "main") |
| 307 | head_commit_id = str(data.get("head") or "") |
| 308 | commits_count = int(data.get("commits_count") or 0) |
| 309 | blobs_count = int(data.get("blobs_count") or 0) |
| 310 | force = bool(data.get("force") or False) |
| 311 | _settings = get_settings() |
| 312 | if commits_count > _settings.mpack_max_commits: |
| 313 | raise HTTPException( |
| 314 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 315 | detail=f"commits_count {commits_count:,} exceeds limit {_settings.mpack_max_commits:,}", |
| 316 | ) |
| 317 | if blobs_count > _settings.mpack_max_objects: |
| 318 | raise HTTPException( |
| 319 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 320 | detail=f"blobs_count {blobs_count:,} exceeds limit {_settings.mpack_max_objects:,}", |
| 321 | ) |
| 322 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 323 | from musehub.services.musehub_wire import NonFastForwardError as _NonFastForwardError |
| 324 | try: |
| 325 | result = await wire_push_unpack_mpack( |
| 326 | session, repo_id, mpack_key, claims.handle, |
| 327 | branch=branch, head_commit_id=head_commit_id, |
| 328 | commits_count=commits_count, blobs_count=blobs_count, |
| 329 | force=force, |
| 330 | ) |
| 331 | except _NonFastForwardError as exc: |
| 332 | raise HTTPException( |
| 333 | status_code=status.HTTP_409_CONFLICT, |
| 334 | detail=str(exc), |
| 335 | ) |
| 336 | except ValueError as exc: |
| 337 | raise HTTPException( |
| 338 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 339 | detail=str(exc), |
| 340 | ) |
| 341 | |
| 342 | # Enqueue intel + file-last-commits + gc + profile.snapshot after every push. |
| 343 | try: |
| 344 | from musehub.services.musehub_jobs import enqueue_push_intel as _enqueue_push_intel |
| 345 | repo_row = await session.get(MusehubRepo, repo_id) |
| 346 | domain_id = repo_row.domain_id if repo_row else None |
| 347 | head = result.get("head", head_commit_id) |
| 348 | await _enqueue_push_intel( |
| 349 | session, |
| 350 | repo_id, |
| 351 | head, |
| 352 | domain_id=domain_id, |
| 353 | branch=branch, |
| 354 | owner=owner, |
| 355 | mpack_key=mpack_key, |
| 356 | ) |
| 357 | await session.commit() |
| 358 | except Exception: |
| 359 | logger.exception( |
| 360 | "enqueue_push_intel failed for repo=%s — push succeeded, intel skipped", |
| 361 | repo_id[:16], |
| 362 | ) |
| 363 | |
| 364 | return _mpack_response(result, request) |
| 365 | |
| 366 | |
| 367 | @router.post( |
| 368 | "/{owner}/{slug}/repair-object", |
| 369 | summary="Replace a corrupt stored object with verified correct bytes (owner/write only)", |
| 370 | status_code=status.HTTP_200_OK, |
| 371 | ) |
| 372 | @limiter.limit(REPAIR_LIMIT) |
| 373 | async def repair_object( |
| 374 | request: Request, |
| 375 | owner: SlugParam, |
| 376 | slug: SlugParam, |
| 377 | claims: TokenClaims = Depends(require_valid_token), |
| 378 | session: AsyncSession = Depends(get_session), |
| 379 | ) -> Response: |
| 380 | """Replace a stored object's bytes with correct content, verified by SHA-256. |
| 381 | |
| 382 | Intended for operators to repair objects that were stored with wrong bytes — |
| 383 | e.g. objects produced by a failed delta reconstruction where the base was |
| 384 | zlib-compressed and the delta result is garbage. |
| 385 | |
| 386 | Request body (msgpack): |
| 387 | object_id str — full "sha256:<64-hex>" content ID |
| 388 | content bytes — the correct raw bytes for this object |
| 389 | |
| 390 | The endpoint verifies SHA-256(content) == object_id before writing. |
| 391 | Only the repo owner or a write/admin collaborator may call this. |
| 392 | """ |
| 393 | raw = await request.body() |
| 394 | ct = request.headers.get("Content-Type", "") |
| 395 | data = _decode_request_body(raw, ct) |
| 396 | object_id: str = data.get("object_id", "") |
| 397 | content: bytes = data.get("content", b"") |
| 398 | if not object_id or not isinstance(content, bytes): |
| 399 | raise HTTPException( |
| 400 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 401 | detail="repair-object requires object_id (str) and content (bytes)", |
| 402 | ) |
| 403 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 404 | caller_id: str | None = claims.handle |
| 405 | try: |
| 406 | result = await wire_repair_object(session, repo_id, object_id, content, caller_id) |
| 407 | except PermissionError as exc: |
| 408 | raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail=str(exc)) |
| 409 | except ObjectHashMismatch as exc: |
| 410 | raise HTTPException(status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, detail=str(exc)) |
| 411 | except ValueError as exc: |
| 412 | raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(exc)) |
| 413 | return _mpack_response(result, request) |
| 414 | |
| 415 | @router.post( |
| 416 | "/{owner}/{slug}/repair-snapshot", |
| 417 | summary="Replace a corrupt snapshot manifest with verified correct content (owner/write only)", |
| 418 | status_code=status.HTTP_200_OK, |
| 419 | ) |
| 420 | @limiter.limit(REPAIR_LIMIT) |
| 421 | async def repair_snapshot( |
| 422 | request: Request, |
| 423 | owner: SlugParam, |
| 424 | slug: SlugParam, |
| 425 | claims: TokenClaims = Depends(require_valid_token), |
| 426 | session: AsyncSession = Depends(get_session), |
| 427 | ) -> Response: |
| 428 | """Replace a stored snapshot's manifest with correct content, verified by snapshot_id. |
| 429 | |
| 430 | Intended for operators to repair snapshots that were stored with an empty |
| 431 | or corrupted manifest_blob — e.g. snapshots affected by the R2 empty-object |
| 432 | bug. Uses force-overwrite (unlike the push path which uses ON CONFLICT DO |
| 433 | NOTHING), so an existing row with wrong content is corrected in place. |
| 434 | |
| 435 | Request body (msgpack): |
| 436 | snapshot_id str — full "sha256:<64-hex>" snapshot ID |
| 437 | manifest dict[str,str] — {path: object_id} mapping |
| 438 | directories list[str] — directory paths (may be empty) |
| 439 | |
| 440 | The endpoint recomputes compute_snapshot_id(manifest, directories) and |
| 441 | verifies it matches snapshot_id before writing. |
| 442 | Only the repo owner or a write/admin collaborator may call this. |
| 443 | """ |
| 444 | raw = await request.body() |
| 445 | ct = request.headers.get("Content-Type", "") |
| 446 | data = _decode_request_body(raw, ct) |
| 447 | snapshot_id: str = data.get("snapshot_id", "") |
| 448 | manifest: JSONObject = data.get("manifest", {}) |
| 449 | directories: list[str] = data.get("directories", []) |
| 450 | if not snapshot_id or not isinstance(manifest, dict): |
| 451 | raise HTTPException( |
| 452 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 453 | detail="repair-snapshot requires snapshot_id (str) and manifest (dict)", |
| 454 | ) |
| 455 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 456 | caller_id: str | None = claims.handle |
| 457 | try: |
| 458 | result = await wire_repair_snapshot( |
| 459 | session, repo_id, snapshot_id, manifest, directories, caller_id |
| 460 | ) |
| 461 | except PermissionError as exc: |
| 462 | raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail=str(exc)) |
| 463 | except ObjectHashMismatch as exc: |
| 464 | raise HTTPException(status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, detail=str(exc)) |
| 465 | except ValueError as exc: |
| 466 | raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(exc)) |
| 467 | return _mpack_response(result, request) |
| 468 | |
| 469 | @router.post( |
| 470 | "/{owner}/{slug}/repair-commit", |
| 471 | summary="Replace a corrupt commit record with verified-correct content (owner/write only)", |
| 472 | status_code=status.HTTP_200_OK, |
| 473 | ) |
| 474 | @limiter.limit(REPAIR_LIMIT) |
| 475 | async def repair_commit( |
| 476 | request: Request, |
| 477 | owner: SlugParam, |
| 478 | slug: SlugParam, |
| 479 | claims: TokenClaims = Depends(require_valid_token), |
| 480 | session: AsyncSession = Depends(get_session), |
| 481 | ) -> Response: |
| 482 | """Replace a stored commit's identity fields with correct content, verified by commit_id. |
| 483 | |
| 484 | Intended for operators to repair commits whose stored row no longer reproduces its |
| 485 | commit_id — e.g. commits the rc10 object-store migration stamped with a |
| 486 | signer_public_key without recomputing the id, so the serve path's hash check fails |
| 487 | on clone and the commit (and all descendants) is dropped. |
| 488 | |
| 489 | Request body (msgpack): |
| 490 | commit dict — a wire commit record (WireCommit shape). Its identity fields |
| 491 | (parent ids, snapshot_id, message, committed_at, author, |
| 492 | signer_public_key) must reproduce commit_id. |
| 493 | |
| 494 | The endpoint recomputes the commit identity (round-tripping committed_at exactly as |
| 495 | the serve path does) and verifies it matches commit_id before writing. |
| 496 | Only the repo owner or a write/admin collaborator may call this. |
| 497 | """ |
| 498 | raw = await request.body() |
| 499 | ct = request.headers.get("Content-Type", "") |
| 500 | data = _decode_request_body(raw, ct) |
| 501 | commit: JSONObject = data.get("commit", {}) |
| 502 | if not isinstance(commit, dict) or not commit.get("commit_id"): |
| 503 | raise HTTPException( |
| 504 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 505 | detail="repair-commit requires commit (dict) with a commit_id", |
| 506 | ) |
| 507 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 508 | caller_id: str | None = claims.handle |
| 509 | try: |
| 510 | result = await wire_repair_commit(session, repo_id, commit, caller_id) |
| 511 | except PermissionError as exc: |
| 512 | raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail=str(exc)) |
| 513 | except ObjectHashMismatch as exc: |
| 514 | raise HTTPException(status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, detail=str(exc)) |
| 515 | except ValueError as exc: |
| 516 | raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(exc)) |
| 517 | return _mpack_response(result, request) |
| 518 | |
| 519 | @router.post( |
| 520 | "/{owner}/{slug}/fetch/objects", |
| 521 | summary="Fetch raw content for a list of object IDs", |
| 522 | status_code=status.HTTP_200_OK, |
| 523 | ) |
| 524 | @limiter.limit(OBJECT_LIMIT) |
| 525 | async def fetch_objects( |
| 526 | request: Request, |
| 527 | owner: SlugParam, |
| 528 | slug: SlugParam, |
| 529 | _claims: TokenClaims | None = Depends(optional_token), |
| 530 | session: AsyncSession = Depends(get_session), |
| 531 | ) -> Response: |
| 532 | """Return raw bytes for each requested object ID as concatenated msgpack frames. |
| 533 | |
| 534 | Request body (msgpack): |
| 535 | object_ids list[str] — canonical sha256:<hex> IDs to fetch |
| 536 | |
| 537 | Response body: concatenated msgpack dicts, one per found object: |
| 538 | {"object_id": str, "content": bytes} |
| 539 | |
| 540 | Objects not found are silently omitted. |
| 541 | """ |
| 542 | raw = await request.body() |
| 543 | ct = request.headers.get("Content-Type", "") |
| 544 | data = _decode_request_body(raw, ct) |
| 545 | object_ids = data.get("object_ids", []) |
| 546 | if not isinstance(object_ids, list): |
| 547 | raise HTTPException( |
| 548 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 549 | detail="fetch/objects requires object_ids (list)", |
| 550 | ) |
| 551 | repo = await _resolve_repo(session, owner, slug) |
| 552 | await _assert_readable(repo, _claims, session) |
| 553 | objects = await wire_fetch_objects(session, repo.repo_id, [str(o) for o in object_ids]) |
| 554 | import msgpack as _mp |
| 555 | body = b"".join(_mp.packb(obj, use_bin_type=True) for obj in objects) |
| 556 | return Response(content=body, media_type="application/x-msgpack") |
| 557 | |
| 558 | @router.post( |
| 559 | "/{owner}/{slug}/fetch/presign", |
| 560 | summary="Wire — presigned fetch for large repos", |
| 561 | status_code=status.HTTP_200_OK, |
| 562 | ) |
| 563 | @limiter.limit(WIRE_FETCH_LIMIT) |
| 564 | async def fetch_presign( |
| 565 | request: Request, |
| 566 | owner: SlugParam, |
| 567 | slug: SlugParam, |
| 568 | _claims: TokenClaims | None = Depends(optional_token), |
| 569 | session: AsyncSession = Depends(get_session), |
| 570 | ) -> Response: |
| 571 | """Return per-object presigned GET URLs for a large fetch delta. |
| 572 | |
| 573 | For large fetches (≥ 500 objects or ≥ 50 MB) Cloudflare's origin timeout |
| 574 | kills the response before the client receives everything. This |
| 575 | endpoint generates one presigned R2 GET URL per needed object — zero object |
| 576 | bytes are read server-side. The client downloads all objects in parallel |
| 577 | directly from R2, bypassing Cloudflare entirely. |
| 578 | |
| 579 | **Request body** (``Content-Type: application/x-msgpack``): |
| 580 | msgpack ``{"want": [str], "have": [str], "depth": int|null, "ttl_seconds": int}`` |
| 581 | |
| 582 | **Response body** (``Content-Type: application/x-msgpack``): |
| 583 | msgpack ``{"presign": bool, "blob_urls": {oid: url}, "commits": [...], |
| 584 | "snapshots": [...], "branch_heads": {...}, "repo_id": str, |
| 585 | "domain": str, "default_branch": str, |
| 586 | "expires_at": str|null, "commit_count": int, "blob_count": int}`` |
| 587 | |
| 588 | When ``presign=False`` the delta is below the threshold or the backend does |
| 589 | not support presigned URLs — the client should fall back to ``fetch/mpack``. |
| 590 | """ |
| 591 | from musehub.services.musehub_wire import wire_fetch_presign |
| 592 | raw = await request.body() |
| 593 | data = _decode_request_body(raw, "application/x-msgpack") |
| 594 | ttl_seconds = int(data.pop("ttl_seconds", 3600)) |
| 595 | body = WireFetchRequest.model_validate(data) |
| 596 | repo = await _resolve_repo(session, owner, slug) |
| 597 | await _assert_readable(repo, _claims, session) |
| 598 | result = await wire_fetch_presign(session, repo.repo_id, body, ttl_seconds) |
| 599 | return _mpack_response(result, request) |
| 600 | |
| 601 | |
| 602 | @router.post( |
| 603 | "/{owner}/{slug}/fetch", |
| 604 | summary="Wire — fetch protocol step 1 (issue #68)", |
| 605 | status_code=status.HTTP_200_OK, |
| 606 | ) |
| 607 | @limiter.limit(WIRE_FETCH_LIMIT) |
| 608 | async def fetch( |
| 609 | request: Request, |
| 610 | owner: SlugParam, |
| 611 | slug: SlugParam, |
| 612 | _claims: TokenClaims | None = Depends(optional_token), |
| 613 | session: AsyncSession = Depends(get_session), |
| 614 | ) -> Response: |
| 615 | """Compute the fetch delta and return a presigned GET URL for the mpack. |
| 616 | |
| 617 | **Request body** (``Content-Type: application/x-msgpack``): |
| 618 | ``{"want": [sha256:...], "have": [sha256:...]}`` |
| 619 | |
| 620 | **Response body** (``Content-Type: application/x-msgpack``): |
| 621 | ``{"mpack_id": "sha256:...", "mpack_url": "...", "commit_count": N, "object_count": N}`` |
| 622 | |
| 623 | Returns 422 if want is empty or entries are malformed. |
| 624 | Returns 404 if any want commit_id is unknown. |
| 625 | Returns 503 with Retry-After if needed objects are not yet indexed. |
| 626 | """ |
| 627 | raw = await request.body() |
| 628 | data = _decode_request_body(raw, "application/x-msgpack") |
| 629 | want: list[str] = data.get("want") or [] |
| 630 | have: list[str] = data.get("have") or [] |
| 631 | |
| 632 | if not want: |
| 633 | raise HTTPException(status_code=422, detail="want must be non-empty") |
| 634 | |
| 635 | repo = await _resolve_repo(session, owner, slug) |
| 636 | await _assert_readable(repo, _claims, session) |
| 637 | |
| 638 | try: |
| 639 | result = await wire_fetch(session, repo.repo_id, want, have) |
| 640 | except MPackValidationError as exc: |
| 641 | raise HTTPException(status_code=422, detail=str(exc)) from exc |
| 642 | except FetchCommitNotFound as exc: |
| 643 | raise HTTPException(status_code=404, detail=str(exc)) from exc |
| 644 | except FetchNotReady as exc: |
| 645 | from fastapi.responses import Response as _Resp |
| 646 | body = msgpack.packb({"error": str(exc), "retry_after": 60}, use_bin_type=True) |
| 647 | return _Resp( |
| 648 | content=body, |
| 649 | status_code=503, |
| 650 | headers={"Retry-After": "60", "Content-Type": "application/x-msgpack"}, |
| 651 | ) |
| 652 | |
| 653 | return _mpack_response(result, request) |
| 654 | |
| 655 | |
| 656 | @router.post( |
| 657 | "/{owner}/{slug}/fetch/mpack", |
| 658 | summary="Wire — single-mpack fetch (issue #47)", |
| 659 | status_code=status.HTTP_200_OK, |
| 660 | ) |
| 661 | @limiter.limit(WIRE_FETCH_LIMIT) |
| 662 | async def fetch_mpack( |
| 663 | request: Request, |
| 664 | owner: SlugParam, |
| 665 | slug: SlugParam, |
| 666 | _claims: TokenClaims | None = Depends(optional_token), |
| 667 | session: AsyncSession = Depends(get_session), |
| 668 | ) -> Response: |
| 669 | """Return the fetch delta as a single content-addressed mpack. |
| 670 | |
| 671 | Phase 2 protocol: server assembles the fetch mpack, stores it ephemerally |
| 672 | in MinIO, and returns a presigned GET URL. The client GETs the mpack |
| 673 | directly from MinIO, verifies sha256, then calls apply_mpack(). |
| 674 | |
| 675 | **Request body** (``Content-Type: application/x-msgpack``): |
| 676 | msgpack ``{"want": [str], "have": [str], "ttl_seconds": int}`` |
| 677 | |
| 678 | **Response body** (``Content-Type: application/x-msgpack``): |
| 679 | msgpack ``{"mpack_url": str|null, "mpack_id": str|null, |
| 680 | "commit_count": int, "blob_count": int}`` |
| 681 | |
| 682 | Returns 503 with ``Retry-After: 60`` if needed objects are not yet indexed |
| 683 | (mpack.index background job still running). |
| 684 | """ |
| 685 | raw = await request.body() |
| 686 | data = _decode_request_body(raw, "application/x-msgpack") |
| 687 | ttl_seconds = int(data.pop("ttl_seconds", 3600)) |
| 688 | want: list[str] = data.get("want") or [] |
| 689 | have: list[str] = data.get("have") or [] |
| 690 | repo = await _resolve_repo(session, owner, slug) |
| 691 | await _assert_readable(repo, _claims, session) |
| 692 | try: |
| 693 | result = await wire_fetch_mpack(session, repo.repo_id, want, have, ttl_seconds) |
| 694 | except MPackNotReadyError: |
| 695 | return Response( |
| 696 | content="mpack not ready — prebuild in progress. Retry shortly.", |
| 697 | status_code=503, |
| 698 | headers={"Retry-After": "30"}, |
| 699 | ) |
| 700 | except FetchNotIndexedError as exc: |
| 701 | return Response( |
| 702 | content=f"Objects not yet indexed: {exc.missing_count} missing. Retry shortly.", |
| 703 | status_code=503, |
| 704 | headers={"Retry-After": "60"}, |
| 705 | ) |
| 706 | return _mpack_response(result, request) |
| 707 | |
| 708 | |
| 709 | # ── release wire endpoints ───────────────────────────────────────────────────── |
| 710 | |
| 711 | @router.post( |
| 712 | "/{owner}/{slug}/releases", |
| 713 | summary="Push a release from muse CLI", |
| 714 | status_code=status.HTTP_201_CREATED, |
| 715 | ) |
| 716 | @limiter.limit(WIRE_PUSH_LIMIT) |
| 717 | async def wire_create_release( |
| 718 | request: Request, |
| 719 | owner: SlugParam, |
| 720 | slug: SlugParam, |
| 721 | background_tasks: BackgroundTasks, |
| 722 | claims: TokenClaims = Depends(require_valid_token), |
| 723 | session: AsyncSession = Depends(get_session), |
| 724 | ) -> Response: |
| 725 | """Accept a ``ReleaseDict`` payload from ``muse release push``. |
| 726 | |
| 727 | The body must be a JSON object matching the CLI ``ReleaseRecord.to_dict()`` |
| 728 | shape (application/json). Returns immediately with the server-assigned |
| 729 | ``release_id``; semantic analysis runs as a background task so the push |
| 730 | is never blocked by analysis time. |
| 731 | |
| 732 | Only the repo owner may push releases — the endpoint mirrors the auth |
| 733 | model of ``POST /{owner}/{slug}/push``. |
| 734 | """ |
| 735 | from musehub.services import musehub_releases as rel_svc |
| 736 | from musehub.services.release_analysis import analyse_release_background |
| 737 | |
| 738 | raw = await request.body() |
| 739 | ct = request.headers.get("Content-Type", "") |
| 740 | data = _decode_request_body(raw, ct) |
| 741 | |
| 742 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 743 | |
| 744 | try: |
| 745 | response = await rel_svc.create_release_from_dict(session, repo_id, data) |
| 746 | except ValueError as exc: |
| 747 | raise HTTPException( |
| 748 | status_code=status.HTTP_409_CONFLICT, |
| 749 | detail=str(exc), |
| 750 | ) |
| 751 | |
| 752 | await session.commit() |
| 753 | logger.info("✅ wire: release %s pushed for %s/%s", response.tag, owner, slug) |
| 754 | |
| 755 | # Fire semantic analysis after the response is sent. Opens its own |
| 756 | # session so it is independent of the request lifecycle. |
| 757 | background_tasks.add_task( |
| 758 | analyse_release_background, repo_id, response.release_id |
| 759 | ) |
| 760 | |
| 761 | return _mpack_response({"release_id": response.release_id}, request) |
| 762 | |
| 763 | @router.delete( |
| 764 | "/{owner}/{slug}/branches/{branch_name:path}", |
| 765 | summary="Delete a branch from MuseHub", |
| 766 | status_code=status.HTTP_200_OK, |
| 767 | ) |
| 768 | @limiter.limit(WIRE_PUSH_LIMIT) |
| 769 | async def wire_delete_branch( |
| 770 | request: Request, |
| 771 | owner: SlugParam, |
| 772 | slug: SlugParam, |
| 773 | branch_name: str, |
| 774 | claims: TokenClaims = Depends(require_valid_token), |
| 775 | session: AsyncSession = Depends(get_session), |
| 776 | ) -> Response: |
| 777 | """Delete a branch pushed by ``muse push``. |
| 778 | |
| 779 | Idiomatic equivalent of ``git push origin --delete <branch>``. The branch |
| 780 | ref is removed from MuseHub; commits and objects are unaffected. |
| 781 | |
| 782 | Only the repo owner may delete branches. Attempting to delete the |
| 783 | repository's default branch is rejected with 409. |
| 784 | """ |
| 785 | from sqlalchemy import delete as sql_delete |
| 786 | |
| 787 | repo = await _resolve_repo(session, owner, slug) |
| 788 | |
| 789 | caller_id: str | None = claims.handle |
| 790 | if caller_id != repo.owner: |
| 791 | raise HTTPException( |
| 792 | status_code=status.HTTP_403_FORBIDDEN, |
| 793 | detail="only the repo owner may delete branches", |
| 794 | ) |
| 795 | |
| 796 | default_branch: str = getattr(repo, "default_branch", None) or "main" |
| 797 | if branch_name == default_branch: |
| 798 | raise HTTPException( |
| 799 | status_code=status.HTTP_409_CONFLICT, |
| 800 | detail=f"cannot delete the default branch '{default_branch}'", |
| 801 | ) |
| 802 | |
| 803 | result = await session.execute( |
| 804 | sql_delete(MusehubBranch).where( |
| 805 | MusehubBranch.repo_id == repo.repo_id, |
| 806 | MusehubBranch.name == branch_name, |
| 807 | ).returning(MusehubBranch.name) |
| 808 | ) |
| 809 | deleted_name: str | None = result.scalar_one_or_none() |
| 810 | if deleted_name is None: |
| 811 | raise HTTPException( |
| 812 | status_code=status.HTTP_404_NOT_FOUND, |
| 813 | detail=f"branch '{branch_name}' not found", |
| 814 | ) |
| 815 | |
| 816 | await session.commit() |
| 817 | logger.info("✅ wire: branch %s deleted from %s/%s", branch_name, owner, slug) |
| 818 | return _mpack_response({"deleted": branch_name}, request) |
| 819 | |
| 820 | @router.delete( |
| 821 | "/{owner}/{slug}/releases/{tag:path}", |
| 822 | summary="Retract a release from MuseHub", |
| 823 | status_code=status.HTTP_200_OK, |
| 824 | ) |
| 825 | @limiter.limit(WIRE_PUSH_LIMIT) |
| 826 | async def wire_delete_release( |
| 827 | request: Request, |
| 828 | owner: SlugParam, |
| 829 | slug: SlugParam, |
| 830 | tag: str, |
| 831 | claims: TokenClaims = Depends(require_valid_token), |
| 832 | session: AsyncSession = Depends(get_session), |
| 833 | ) -> Response: |
| 834 | """Retract a release pushed by ``muse release push``. |
| 835 | |
| 836 | Removes the named release label from MuseHub. The underlying commits and |
| 837 | snapshots are not affected — they remain in the content-addressed object |
| 838 | store and are still reachable by their SHA-256. |
| 839 | |
| 840 | Only the repo owner may retract releases. |
| 841 | """ |
| 842 | from musehub.services import musehub_releases as rel_svc |
| 843 | |
| 844 | repo = await _resolve_repo(session, owner, slug) |
| 845 | |
| 846 | caller_id: str | None = claims.handle |
| 847 | if caller_id != repo.owner: |
| 848 | raise HTTPException( |
| 849 | status_code=status.HTTP_403_FORBIDDEN, |
| 850 | detail="only the repo owner may retract releases", |
| 851 | ) |
| 852 | |
| 853 | deleted = await rel_svc.delete_release_by_tag(session, repo.repo_id, tag) |
| 854 | if not deleted: |
| 855 | raise HTTPException( |
| 856 | status_code=status.HTTP_404_NOT_FOUND, |
| 857 | detail=f"release '{tag}' not found", |
| 858 | ) |
| 859 | |
| 860 | await session.commit() |
| 861 | logger.info("✅ wire: release %s retracted from %s/%s", tag, owner, slug) |
| 862 | return _mpack_response({"retracted": tag}, request) |
| 863 | |
| 864 | # ── wire-tag endpoints ───────────────────────────────────────────────────────── |
| 865 | |
| 866 | @router.post( |
| 867 | "/{owner}/{slug}/tags", |
| 868 | summary="Push lightweight wire tags from muse CLI", |
| 869 | status_code=status.HTTP_200_OK, |
| 870 | ) |
| 871 | @limiter.limit(WIRE_PUSH_LIMIT) |
| 872 | async def wire_push_tags( |
| 873 | request: Request, |
| 874 | owner: SlugParam, |
| 875 | slug: SlugParam, |
| 876 | claims: TokenClaims = Depends(require_valid_token), |
| 877 | session: AsyncSession = Depends(get_session), |
| 878 | ) -> Response: |
| 879 | """Upsert a batch of lightweight semantic tags pushed from the Muse CLI. |
| 880 | |
| 881 | The body must be a msgpack object with a ``tags`` key holding a list of |
| 882 | ``WireTag`` dicts. The server upserts them — pushing the same tag label |
| 883 | twice for the same repo is a no-op (``commit_id`` is refreshed). |
| 884 | |
| 885 | Returns ``{"stored": <count>}`` as msgpack or JSON. |
| 886 | """ |
| 887 | from musehub.models.musehub import WireTagInput |
| 888 | from musehub.services import musehub_wire_tags as tag_svc |
| 889 | |
| 890 | raw = await request.body() |
| 891 | ct = request.headers.get("Content-Type", "") |
| 892 | try: |
| 893 | data = _decode_request_body(raw, ct) |
| 894 | except (ValueError, Exception): |
| 895 | raise HTTPException( |
| 896 | status_code=status.HTTP_400_BAD_REQUEST, |
| 897 | detail="Request body must be a valid msgpack or JSON object", |
| 898 | ) |
| 899 | |
| 900 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 901 | |
| 902 | tags_raw = data.get("tags", []) |
| 903 | if not isinstance(tags_raw, list): |
| 904 | raise HTTPException( |
| 905 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 906 | detail="'tags' must be a list", |
| 907 | ) |
| 908 | |
| 909 | tags: list[WireTagInput] = [] |
| 910 | for item in tags_raw: |
| 911 | if not isinstance(item, dict): |
| 912 | continue |
| 913 | tag_id_raw = item.get("tag_id", "") |
| 914 | commit_id_raw = item.get("commit_id", "") |
| 915 | tag_label_raw = item.get("tag", "") |
| 916 | created_at_raw = item.get("created_at", "") |
| 917 | tags.append( |
| 918 | WireTagInput( |
| 919 | tag_id=str(tag_id_raw) if isinstance(tag_id_raw, str) else "", |
| 920 | commit_id=str(commit_id_raw) if isinstance(commit_id_raw, str) else "", |
| 921 | tag=str(tag_label_raw) if isinstance(tag_label_raw, str) else "", |
| 922 | created_at=str(created_at_raw) if isinstance(created_at_raw, str) else "", |
| 923 | ) |
| 924 | ) |
| 925 | |
| 926 | stored = await tag_svc.store_wire_tags(session, repo_id, tags) |
| 927 | await session.commit() |
| 928 | logger.info("✅ wire: %d tag(s) pushed for %s/%s", stored, owner, slug) |
| 929 | return _mpack_response({"stored": stored}, request) |
| 930 | |
| 931 | # ── version-tag endpoints ────────────────────────────────────────────────────── |
| 932 | |
| 933 | @router.post( |
| 934 | "/{owner}/{slug}/version-tags", |
| 935 | summary="Push semantic-version tags from muse CLI", |
| 936 | status_code=status.HTTP_200_OK, |
| 937 | ) |
| 938 | @limiter.limit(WIRE_PUSH_LIMIT) |
| 939 | async def wire_push_version_tags( |
| 940 | request: Request, |
| 941 | owner: SlugParam, |
| 942 | slug: SlugParam, |
| 943 | force: bool = False, |
| 944 | claims: TokenClaims = Depends(require_valid_token), |
| 945 | session: AsyncSession = Depends(get_session), |
| 946 | ) -> Response: |
| 947 | """Upsert a batch of semantic-version tags pushed from the Muse CLI. |
| 948 | |
| 949 | The body must be a msgpack object with a ``tags`` key holding a list of |
| 950 | version-tag dicts. Re-pushing the same tag label without ``force=true`` |
| 951 | is a no-op if the commit_id is unchanged (skipped). With ``force=true`` |
| 952 | the commit_id is overwritten. |
| 953 | |
| 954 | Returns ``{"stored": <count>, "skipped": <count>}`` as msgpack or JSON. |
| 955 | Returns 409 if a tag already exists with a different commit_id and |
| 956 | ``force`` is not set. |
| 957 | """ |
| 958 | from musehub.services import musehub_version_tags as vtag_svc |
| 959 | |
| 960 | raw = await request.body() |
| 961 | ct = request.headers.get("Content-Type", "") |
| 962 | try: |
| 963 | data = _decode_request_body(raw, ct) |
| 964 | except (ValueError, Exception): |
| 965 | raise HTTPException( |
| 966 | status_code=status.HTTP_400_BAD_REQUEST, |
| 967 | detail="Request body must be a valid msgpack or JSON object", |
| 968 | ) |
| 969 | |
| 970 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 971 | |
| 972 | tags_raw = data.get("tags", []) |
| 973 | if not isinstance(tags_raw, list): |
| 974 | raise HTTPException( |
| 975 | status_code=status.HTTP_422_UNPROCESSABLE_CONTENT, |
| 976 | detail="'tags' must be a list", |
| 977 | ) |
| 978 | |
| 979 | tags: list[dict] = [] |
| 980 | for item in tags_raw: |
| 981 | if not isinstance(item, dict): |
| 982 | continue |
| 983 | tags.append(item) |
| 984 | |
| 985 | # Without force, detect conflicts (same tag label, different commit_id) first. |
| 986 | if not force: |
| 987 | from sqlalchemy import select as sa_select |
| 988 | from musehub.db.musehub_repo_models import MusehubVersionTag as _VTag |
| 989 | for item in tags: |
| 990 | tag_label = str(item.get("tag", "")).strip() |
| 991 | commit_id = str(item.get("commit_id", "")).strip() |
| 992 | if not tag_label or not commit_id: |
| 993 | continue |
| 994 | existing = await session.scalar( |
| 995 | sa_select(_VTag).where( |
| 996 | _VTag.repo_id == repo_id, |
| 997 | _VTag.tag == tag_label, |
| 998 | ).limit(1) |
| 999 | ) |
| 1000 | if existing is not None and existing.commit_id != commit_id: |
| 1001 | raise HTTPException( |
| 1002 | status_code=status.HTTP_409_CONFLICT, |
| 1003 | detail=f"Tag '{tag_label}' already points to a different commit. Use force=true to override.", |
| 1004 | ) |
| 1005 | |
| 1006 | stored, skipped = await vtag_svc.store_version_tags(session, repo_id, tags, force=force) |
| 1007 | await session.commit() |
| 1008 | logger.info( |
| 1009 | "✅ wire: %d version tag(s) stored, %d skipped for %s/%s", |
| 1010 | stored, skipped, owner, slug, |
| 1011 | ) |
| 1012 | return _mpack_response({"stored": stored, "skipped": skipped}, request) |
| 1013 | |
| 1014 | |
| 1015 | @router.get( |
| 1016 | "/{owner}/{slug}/version-tags", |
| 1017 | summary="Fetch all version tags for a repo", |
| 1018 | status_code=status.HTTP_200_OK, |
| 1019 | ) |
| 1020 | @limiter.limit(OBJECT_LIMIT) |
| 1021 | async def wire_get_version_tags( |
| 1022 | request: Request, |
| 1023 | owner: SlugParam, |
| 1024 | slug: SlugParam, |
| 1025 | _claims: TokenClaims = Depends(require_valid_token), |
| 1026 | session: AsyncSession = Depends(get_session), |
| 1027 | ) -> Response: |
| 1028 | """Return all semantic-version tags stored for a repo. |
| 1029 | |
| 1030 | Returns ``{"tags": [...]}`` as msgpack or JSON, with tags ordered by |
| 1031 | semver descending (newest first). |
| 1032 | """ |
| 1033 | from musehub.services import musehub_version_tags as vtag_svc |
| 1034 | |
| 1035 | repo_id = await _resolve_repo_id(session, owner, slug) |
| 1036 | tags = await vtag_svc.list_version_tags(session, repo_id) |
| 1037 | logger.info("✅ wire: returning %d version tag(s) for %s/%s", len(tags), owner, slug) |
| 1038 | return _mpack_response({"tags": tags}, request) |
| 1039 | |
| 1040 | # ── content-addressed CDN ────────────────────────────────────────────────────── |
| 1041 | |
| 1042 | @router.get( |
| 1043 | "/o/{object_id:path}", |
| 1044 | summary="Content-addressed object CDN endpoint", |
| 1045 | response_description="Raw binary blob", |
| 1046 | tags=["Objects"], |
| 1047 | ) |
| 1048 | @limiter.limit(OBJECT_LIMIT) |
| 1049 | async def get_object( |
| 1050 | request: Request, |
| 1051 | object_id: str, |
| 1052 | repo_id: str | None = None, |
| 1053 | _claims: TokenClaims | None = Depends(optional_token), |
| 1054 | session: AsyncSession = Depends(get_session), |
| 1055 | ) -> Response: |
| 1056 | """Serve a content-addressed binary object. |
| 1057 | |
| 1058 | Objects are immutable (ID is derived from content hash), so the response |
| 1059 | carries ``Cache-Control: max-age=31536000, immutable`` — safe to place |
| 1060 | behind CloudFront forever. |
| 1061 | """ |
| 1062 | from musehub.storage.backends import read_object_bytes |
| 1063 | backend = get_backend() |
| 1064 | try: |
| 1065 | raw = await backend.get(object_id) |
| 1066 | except ValueError: |
| 1067 | raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="invalid object path") |
| 1068 | if raw is None: |
| 1069 | obj_row = await session.scalar( |
| 1070 | select(MusehubObject).where(MusehubObject.object_id == object_id).limit(1) |
| 1071 | ) |
| 1072 | if obj_row is not None: |
| 1073 | raw = await read_object_bytes(obj_row, session=session) |
| 1074 | if raw is None: |
| 1075 | logger.error( |
| 1076 | "OBJECT_404 object_id=%s repo_id=%s backend=%s", |
| 1077 | object_id, |
| 1078 | repo_id, |
| 1079 | type(get_backend()).__name__, |
| 1080 | ) |
| 1081 | raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="object not found") |
| 1082 | |
| 1083 | # Integrity check: verify stored bytes match the content-addressed ID. |
| 1084 | import hashlib as _hashlib |
| 1085 | actual_hash = _hashlib.sha256(raw).hexdigest() |
| 1086 | expected_hex = object_id.removeprefix("sha256:") |
| 1087 | if actual_hash != expected_hex: |
| 1088 | logger.error( |
| 1089 | "OBJECT_CORRUPT object_id=%s actual_hash=sha256:%s size=%d backend=%s", |
| 1090 | object_id, |
| 1091 | actual_hash, |
| 1092 | len(raw), |
| 1093 | type(get_backend()).__name__, |
| 1094 | ) |
| 1095 | |
| 1096 | return Response( |
| 1097 | content=raw, |
| 1098 | media_type="application/octet-stream", |
| 1099 | headers={ |
| 1100 | "Cache-Control": "public, max-age=31536000, immutable", |
| 1101 | "ETag": f'"{object_id}"', |
| 1102 | }, |
| 1103 | ) |
| 1104 |
File History
4 commits
sha256:92528ae07d0e1239d87fd5fd1f439e8fbb49c9778a9a400bc4a736073fb28316
feat: byte-range blob reads, file attribution DAG walk, bra…
Sonnet 4.6
minor
⚠
43 days ago
sha256:e652528bc7ab28fc0a6799df687779ee296be645a7e1b4120e271df96aebf20a
chore: carry uncommitted changes (mpack_key in enqueue_push…
Sonnet 4.6
minor
⚠
45 days ago
sha256:ab9eda7b6479e1c35cdba9a54f62bacd2825de8faacec3ba67a9a8ef45914b7d
fix: migration and wire protocol alignment
Sonnet 4.6
minor
⚠
48 days ago
sha256:ef10830ce231e0a20efcb0e2586cb879471247e916616e6fdd0d51df459e2595
fix: typing audit — 0 violations, 0 untyped defs across all…
Sonnet 4.6
minor
⚠
49 days ago