fix: an unfinished download is never stored as an attachment, and the temp sweep keeps any file a library row points at

The importer refuses a staging file (.part, .partial, .megatmp.*) at both
entry points, the same way it refuses a sidecar .json, and
discard_after_attach never deletes one: the age-gated temp sweep owns them.
The sweep checks its stale candidates against image and attachment rows and
keeps any a row points at.

A Google Drive fetch's half-written file had been stored as <sha>.part; the
sweep then deleted it by name and left the row with no file.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

backend/app/services/importer.py

@@ -43,6 +43,7 @@ from ..utils.paths import (
 from ..utils.phash import compute_phash, find_similar, fingerprint_path, fingerprints_match
 from ..utils.sidecar import find_sidecar, parse_sidecar
 from ..utils.slug import slugify
+from ..utils.staging import is_temp_staging_file
 from .archive_extractor import extract_archive, is_archive
 from .attachment_store import AttachmentStore
 from .audits import single_color
@@ -115,6 +116,21 @@ def is_video(path: Path) -> bool:
     return path.suffix.lower() in VIDEO_EXTS
 
 
+def _not_content(path: Path) -> ImportResult | None:
+    """The skip result for a file that is never library content: a sidecar
+    `.json` (metadata) or a downloader's staging file (an unfinished download,
+    which becomes content only under its final name). None for anything else."""
+    if path.suffix.lower() == ".json":
+        error = "sidecar json is metadata, not content"
+    elif is_temp_staging_file(path):
+        error = "unfinished download (staging file), not content"
+    else:
+        return None
+    return ImportResult(
+        status="skipped", skip_reason=SkipReason.invalid_image, error=error,
+    )
+
+
 def _safe_ext(path: Path) -> str:
     """Conservatively extract a file extension for PostAttachment.ext
     (varchar(32)). Thin wrapper over the shared `utils.paths.safe_ext` (kept for
@@ -410,11 +426,8 @@ class Importer:
         media members (one Post via the archive-adjacent sidecar) and
         preserve the archive itself. Other non-media → PostAttachment.
         Nothing is skipped for being non-media (FC-2d-iii)."""
-        if source.suffix.lower() == ".json":
-            return ImportResult(
-                status="skipped", skip_reason=SkipReason.invalid_image,
-                error="sidecar json is metadata, not content",
-            )
+        if (refused := _not_content(source)) is not None:
+            return refused
         if is_archive(source):
             return self._import_archive(source)
         if not is_supported(source):
@@ -1219,11 +1232,8 @@ class Importer:
             caller may delete the on-disk original (mirrors duplicate_hash)
           - failed → file untouched; caller decides
         """
-        if path.suffix.lower() == ".json":
-            return ImportResult(
-                status="skipped", skip_reason=SkipReason.invalid_image,
-                error="sidecar json is metadata, not content",
-            )
+        if (refused := _not_content(path)) is not None:
+            return refused
         if is_archive(path):
             return self._import_archive(
                 path, artist=artist, source_row=source,
@@ -1420,7 +1430,11 @@ class Importer:
 
         An imported or superseded media file is kept: it IS the record's file.
         So is any path an ImageRecord points at, whatever the status, because a
-        re-fetch can write identical bytes over a record's own file."""
+        re-fetch can write identical bytes over a record's own file. A staging
+        file is kept too: it may be a download still being written, and the
+        temp-file sweep removes it once it is old enough to be abandoned."""
+        if is_temp_staging_file(path):
+            return False
         if result.status in ("imported", "superseded"):
             redundant = is_archive(path)
         else:

backend/app/tasks/maintenance.py

@@ -20,6 +20,7 @@ from ..models import (
     ImportSettings,
     ImportTask,
     LibraryAuditRun,
+    PostAttachment,
     Source,
     TaskRun,
 )
@@ -372,23 +373,58 @@ def cleanup_old_tasks() -> int:
         return result.rowcount or 0
 
 
+def _paths_with_rows(paths: list[str]) -> set[str]:
+    """The subset of `paths` that an image (its file or thumbnail) or an
+    attachment row points at."""
+    owned: set[str] = set()
+    SessionLocal = _sync_session_factory()
+    with SessionLocal() as session:
+        for i in range(0, len(paths), 500):
+            chunk = paths[i:i + 500]
+            owned.update(session.execute(
+                select(ImageRecord.path).where(ImageRecord.path.in_(chunk))
+            ).scalars())
+            owned.update(session.execute(
+                select(ImageRecord.thumbnail_path).where(ImageRecord.thumbnail_path.in_(chunk))
+            ).scalars())
+            owned.update(session.execute(
+                select(PostAttachment.path).where(PostAttachment.path.in_(chunk))
+            ).scalars())
+    return owned
+
+
 @celery.task(name="backend.app.tasks.maintenance.cleanup_orphaned_temp_files")
 def cleanup_orphaned_temp_files() -> int:
     """Delete orphaned staging files (.part, .partial, megatools' .megatmp.*)
     under the images root, left by a download/import killed mid-write. Only
     removes files older than ORPHAN_TEMP_MIN_AGE_HOURS so an in-flight
-    download's staging file is never pulled out from under it. Returns the
-    count removed."""
+    download's staging file is never pulled out from under it, and never a
+    file an image or attachment row points at, whatever its name: that file
+    is library content. Returns the count removed."""
     if not IMAGES_ROOT.is_dir():
         return 0
     cutoff = datetime.now(UTC).timestamp() - ORPHAN_TEMP_MIN_AGE_HOURS * 3600
-    removed = 0
+    stale: list[Path] = []
     for path in IMAGES_ROOT.rglob("*"):
         if not is_temp_staging_file(path) or not path.is_file():
             continue
         try:
-            if path.stat().st_mtime >= cutoff:
-                continue          # too fresh — may be an active download
+            if path.stat().st_mtime < cutoff:   # fresher may be an active download
+                stale.append(path)
+        except OSError as exc:
+            log.warning("cleanup_orphaned_temp_files: %s: %s", path, exc)
+    if not stale:
+        return 0
+    owned = _paths_with_rows([str(p) for p in stale])
+    removed = 0
+    for path in stale:
+        if str(path) in owned:
+            log.warning(
+                "cleanup_orphaned_temp_files: keeping %s: a library row points at it",
+                path,
+            )
+            continue
+        try:
             path.unlink()
             removed += 1
         except OSError as exc:

tests/test_importer_attach_in_place.py

@@ -410,6 +410,31 @@ def test_attach_in_place_invalid_image_returns_skipped(importer):
     assert result.skip_reason.value == "invalid_image"
 
 
+@pytest.mark.parametrize("name", ["film.mp4abc123_.part", "img.jpg.partial", ".megatmp.Xy12"])
+def test_an_unfinished_download_is_refused_and_left_on_disk(importer, db_sync, name):
+    """A staging file is a download still being written (or abandoned): it is
+    never stored as an attachment, and the importer never deletes it, so a
+    download in progress is not pulled out from under its writer."""
+    from backend.app.models import PostAttachment
+
+    artist = Artist(name="Hal", slug="hal")
+    db_sync.add(artist)
+    db_sync.flush()
+    staging = importer.images_root / "hal" / "patreon" / "post" / name
+    staging.parent.mkdir(parents=True, exist_ok=True)
+    staging.write_bytes(b"half of a video")
+
+    for result in (importer.attach_in_place(staging, artist=artist),
+                   importer.import_one(staging)):
+        assert result.status == "skipped"
+        assert "unfinished download" in result.error
+        assert importer.discard_after_attach(staging, result) is False
+    assert staging.exists()
+    assert db_sync.execute(
+        select(func.count()).where(PostAttachment.original_filename == name)
+    ).scalar() == 0
+
+
 def test_attach_in_place_non_media_routes_to_attachment(importer, db_sync):
     """FC-2d-iii dispatch parity for the download path. A non-media file
     (.pdf, .txt etc.) downloaded by gallery-dl must become a PostAttachment,

tests/test_maintenance.py

@@ -26,30 +26,50 @@ def _make_batch(session) -> int:
     return batch.id
 
 
-def test_cleanup_orphaned_temp_files_removes_stale_only(tmp_path, monkeypatch):
+def test_cleanup_orphaned_temp_files_removes_stale_only(db_sync, tmp_path, monkeypatch):
     import os
 
+    from backend.app.models import ImageRecord, PostAttachment
     from backend.app.tasks import maintenance as m
 
     monkeypatch.setattr(m, "IMAGES_ROOT", tmp_path)
-    stale = tmp_path / "artist" / "img.jpg.part"      # killed download → orphan
-    stale.parent.mkdir(parents=True)
-    stale.write_bytes(b"x")
     old = datetime.now(UTC).timestamp() - 8 * 3600     # older than the 6h guard
-    os.utime(stale, (old, old))
-    fresh = tmp_path / "in_progress.jpg.partial"       # active download → keep
-    fresh.write_bytes(b"x")
-    keep = tmp_path / "real.jpg"                        # a real image → keep
+
+    def staged(rel: str, *, aged: bool = True):
+        p = tmp_path / rel
+        p.parent.mkdir(parents=True, exist_ok=True)
+        p.write_bytes(b"x")
+        if aged:
+            os.utime(p, (old, old))
+        return p
+
+    stale = staged("artist/img.jpg.part")                 # killed download → orphan
+    mega = staged("artist/.megatmp.AbCd123")              # abandoned megatools stage
+    fresh = staged("in_progress.jpg.partial", aged=False)  # active download → keep
+    keep = tmp_path / "real.jpg"                          # a real image → keep
     keep.write_bytes(b"x")
-    mega = tmp_path / "artist" / ".megatmp.AbCd123"     # abandoned megatools stage
-    mega.write_bytes(b"x")
-    os.utime(mega, (old, old))
+    # Staging-named files a library row points at are library content.
+    attached = staged("attachments/ab/abc.part")
+    imaged = staged("artist/video.mp4.part")
+    db_sync.add(PostAttachment(
+        post_id=None, artist_id=None, sha256="tempsweep-att".ljust(64, "x"),
+        path=str(attached), original_filename="film.mp4.part", ext=".part",
+        size_bytes=1,
+    ))
+    db_sync.add(ImageRecord(
+        path=str(imaged), sha256="tempsweep-img".ljust(64, "x"), size_bytes=1,
+        mime="video/mp4", width=1, height=1, origin="downloaded",
+        integrity_status="ok",
+    ))
+    db_sync.commit()
 
     assert m.cleanup_orphaned_temp_files() == 2
     assert not stale.exists()
     assert not mega.exists()
     assert fresh.exists()
     assert keep.exists()
+    assert attached.exists()
+    assert imaged.exists()
 
 
 def test_recover_interrupted_only_old(db_sync, monkeypatch):