fix(files): review round 1, the server — the tray's window, the GC's locks, a file's origin, the upgrade path

- The queue keeps a done job by when it became done (ready_at), not by its last write: renaming,
  keeping or moving an old file, or the GC's expiry warning, no longer brings it back as just finished.
- GC expiry and eviction re-read each file under the asset→job locks the user routes take, and commit
  one file at a time: a delete, a re-download or a keep made while the GC runs is honoured, never a
  second release that frees a file another user still holds.
- A file records the origin it was placed under (download_jobs.source_key, migration 0105): an edit
  whose parent expired still goes back to its folder, not to Uploads.
- A folder a file is being put in is held (FOR SHARE) until it lands; a delete of it waits and takes
  that file home too, instead of failing on the foreign key.
- The hash route reads at most two files at once (429 past that) and holds no connection meanwhile.
- 0104's backfill dates old files by their request date only: no notice prod wrote names its job.
- The old DOWNLOAD_ROOT still finds the files; a files root that is not a mounted volume stops the
  worker and is logged by the API; the volume-copy recipe looks the project-prefixed names up.
- Stale comments about the /downloads mount and the Downloads page.

backend/tests/test_file_folders.py

@@ -15,6 +15,7 @@ from app import sysconfig
 from app.config import settings
 from app.downloads import folders, formats, service
 from app.downloads import gc as gcmod
+from app.db import SessionLocal
 from app.downloads.status import JobStatus
 from app.models import DownloadJob, FileFolder, MediaAsset, Notification
 
@@ -123,6 +124,68 @@ def test_an_edit_whose_parent_is_gone_follows_the_parents_asset(db, make_user):
     assert folders.source_key_for(db, edit) == "site:vimeo.com"
 
 
+def test_an_edit_keeps_its_origin_after_its_parent_and_its_file_are_gone(db, make_user):
+    # The parent download expires, its file goes with its last holder, and both of the edit's
+    # pointers to it are cleared: the edit must still go back to the YouTube folder it came from.
+    uid = make_user().id
+    parent = _job(db, uid, _asset(db))
+    edit = _job(
+        db, uid, _asset(db, "edit", "e1"), kind="edit", source_job_id=parent.id,
+        source_asset_id=parent.asset_id,
+    )  # fmt: skip
+    home = _folder(db, edit)
+    own = FileFolder(user_id=uid, kind=folders.FolderKind.OWN, name="Mine")
+    db.add(own)
+    db.flush()
+    folders.move(db, edit, own)
+    edit.source_job_id = edit.source_asset_id = None  # what ON DELETE SET NULL leaves
+    db.delete(parent)
+    db.commit()
+    folders.move(db, edit, home)
+    assert edit.folder_id == home.id
+    folders.move(db, edit, own)
+    folders.delete_own(db, own)
+    assert edit.folder_id == home.id
+
+
+def test_a_folder_deleted_while_a_file_goes_in_takes_that_file_home_too(db, make_user):
+    # Before, the delete missed the file still on its way in and failed on the foreign key (a 500),
+    # or the file's insert did after the folder went.
+    import threading
+
+    uid = make_user().id
+    own = FileFolder(user_id=uid, kind=folders.FolderKind.OWN, name="Mine")
+    db.add(own)
+    db.commit()
+    own_id = own.id
+    job = _job(db, uid, _asset(db))
+    home = job.folder_id
+    target = folders.own_folder(db, uid, own_id)  # held from the check until this commits
+    outcome: list[object] = []
+
+    def delete_it():
+        other = SessionLocal()
+        try:
+            outcome.append(folders.delete_own(other, other.get(FileFolder, own_id)))
+            other.commit()
+        except Exception as e:  # noqa: BLE001 — the assertion below names what went wrong
+            outcome.append(e)
+        finally:
+            other.close()
+
+    t = threading.Thread(target=delete_it)
+    t.start()
+    t.join(timeout=1)
+    assert t.is_alive()  # the delete waits for the file, even before it is written
+    folders.move(db, job, target)
+    db.commit()
+    t.join(timeout=10)
+    assert outcome == [1]
+    db.expire_all()
+    assert db.get(DownloadJob, job.id).folder_id == home
+    assert db.get(FileFolder, own_id) is None
+
+
 def test_folders_are_per_user(db, make_user):
     a, b = make_user().id, make_user().id
     asset = _asset(db)
@@ -201,6 +264,30 @@ def test_the_gc_warns_once_then_takes_the_file_out_of_the_library(db, make_user,
     assert db.execute(select(func.count()).select_from(MediaAsset)).scalar() == 0
 
 
+@pytest.mark.parametrize("restart", ["unkeep", "move", "retention"])
+def test_a_file_whose_clock_starts_again_is_warned_again(db, make_user, _root, restart):
+    # The GC warns once per expiry; a file whose expiry was moved must be warned before the new one.
+    (_root / "a.mp4").write_bytes(b"x")
+    uid = make_user().id
+    job = _job(db, uid, _asset(db), expires_at=NOW + timedelta(hours=12))
+    _gc(db)
+    assert job.expiry_warned
+    if restart == "unkeep":
+        folders.set_kept(db, job, True)
+        folders.set_kept(db, job, False)
+    elif restart == "move":
+        home = _folder(db, job)
+        own = FileFolder(user_id=uid, kind=folders.FolderKind.OWN, name="Mine")
+        db.add(own)
+        db.flush()
+        folders.move(db, job, own)
+        folders.move(db, job, home)
+    else:
+        folders.set_retention(db, _folder(db, job), 90)
+    db.commit()
+    assert not job.expiry_warned
+
+
 def test_the_gc_never_takes_a_kept_file(db, make_user, _root):
     (_root / "a.mp4").write_bytes(b"x")
     uid = make_user().id
@@ -221,6 +308,52 @@ def test_one_users_expiry_leaves_another_users_copy(db, make_user, _root):
     assert (_root / "a.mp4").exists()
 
 
+def _meanwhile(monkeypatch, act):
+    """Run `act` in another session (another request) between the GC's expiry SELECT and its
+    handling of each file it found — the window a user's click can land in."""
+    real = gcmod._relock_expired
+
+    def relock(db, job_id, asset_id, now):
+        other = SessionLocal()
+        try:
+            act(other, job_id)
+            other.commit()
+        finally:
+            other.close()
+        return real(db, job_id, asset_id, now)
+
+    monkeypatch.setattr(gcmod, "_relock_expired", relock)
+
+
+def test_a_file_kept_while_the_gc_runs_stays(db, make_user, _root, monkeypatch):
+    (_root / "a.mp4").write_bytes(b"x")
+    job = _job(db, make_user().id, _asset(db), expires_at=NOW - timedelta(minutes=1))
+    _meanwhile(monkeypatch, lambda other, jid: setattr(other.get(DownloadJob, jid), "kept", True))
+    assert _gc(db)["expired"] == 0
+    db.expire_all()
+    assert db.get(DownloadJob, job.id) is not None
+    assert (_root / "a.mp4").exists()
+
+
+def test_a_file_deleted_while_the_gc_runs_is_released_once(db, make_user, _root, monkeypatch):
+    # Released twice, the shared file would go from under the other user's copy.
+    (_root / "a.mp4").write_bytes(b"x")
+    a, b = make_user().id, make_user().id
+    asset = _asset(db)
+    _job(db, a, asset, expires_at=NOW - timedelta(minutes=1))
+    stays = _job(db, b, asset, expires_at=NOW + timedelta(days=20))
+
+    def user_deletes(other, jid):
+        service.delete(other, other.get(DownloadJob, jid))
+
+    _meanwhile(monkeypatch, user_deletes)
+    _gc(db)
+    db.expire_all()
+    assert db.get(DownloadJob, stays.id).asset_id == asset.id
+    assert db.get(MediaAsset, asset.id).ref_count == 1
+    assert (_root / "a.mp4").exists()
+
+
 def test_the_cache_cap_never_evicts_a_kept_file_and_an_evicted_one_leaves_every_library(
     db, make_user, _root
 ):

backend/tests/test_files_api.py

@@ -421,6 +421,21 @@ def test_a_deleted_file_leaves_the_library_and_the_log_says_so(db, make_user, as
     assert (row.actor_id, row.target_id) == (alice.id, str(job.id))
 
 
+def test_a_file_shared_with_someone_is_theirs_to_watch_not_to_change(db, make_user, as_user, _root):
+    # The read routes let a share's recipient in; the write routes must not.
+    alice, friend = make_user(), make_user()
+    job = _file(db, alice.id, title="Mine")
+    db.add(DownloadShare(job_id=job.id, from_user_id=alice.id, to_user_id=friend.id))
+    db.commit()
+    c = as_user(friend)
+    assert c.patch(f"/api/files/{job.id}", json={"name": "Theirs"}).status_code == 404
+    assert c.patch(f"/api/files/{job.id}", json={"kept": True}).status_code == 404
+    assert c.delete(f"/api/files/{job.id}").status_code == 404
+    db.expire_all()
+    still = db.get(DownloadJob, job.id)
+    assert still is not None and still.display_name is None and not still.kept
+
+
 # --- upload, the file, its hash ------------------------------------------------------------
 
 
@@ -532,6 +547,26 @@ def test_the_file_carries_its_hash_once_known_and_the_hash_is_taken_on_first_ask
     assert as_user(make_user()).get(f"/api/files/{job.id}/sha256").status_code == 404
 
 
+def test_only_a_few_files_are_hashed_at_once(db, make_user, as_user, _root):
+    # Each hash holds a thread of the pool every sync route shares for the whole read; a caller
+    # past the few that may run is told to come back, not queued behind them.
+    from app.routes import files as filesroute
+
+    alice = make_user()
+    job = _file(db, alice.id, rel="v.mp4")
+    (_root / "v.mp4").write_bytes(b"bytes")
+    held = [filesroute._HASHING.acquire(blocking=False) for _ in range(filesroute._HASHERS)]
+    try:
+        assert all(held)
+        busy = as_user(alice).get(f"/api/files/{job.id}/sha256")
+    finally:
+        for _ in held:
+            filesroute._HASHING.release()
+    assert busy.status_code == 429
+    assert busy.json()["detail"]["reason"] == "file_hash_busy"
+    assert as_user(alice).get(f"/api/files/{job.id}/sha256").status_code == 200
+
+
 def test_a_new_file_at_an_assets_path_forgets_the_old_hash(db, make_user):
     job = _file(db, make_user().id)
     asset = db.get(MediaAsset, job.asset_id)
@@ -562,7 +597,7 @@ def test_the_queue_is_what_is_in_flight_and_what_just_finished_not_the_library(
         db.execute(
             update(DownloadJob)
             .where(DownloadJob.id == job.id)
-            .values(updated_at=NOW - timedelta(hours=1))
+            .values(updated_at=NOW - timedelta(hours=1), ready_at=NOW - timedelta(hours=1))
         )
     db.commit()
     rows = as_user(user).get("/api/downloads").json()
@@ -570,3 +605,22 @@ def test_the_queue_is_what_is_in_flight_and_what_just_finished_not_the_library(
     assert {row["id"] for row in rows} == {fresh.id, waiting.id, stuck.id, paused.id, stopped.id}
     # The tray names where each came from, as the library does.
     assert {row["source_key"] for row in rows} == {"youtube"}
+
+
+def test_an_old_file_written_to_does_not_come_back_as_just_finished(db, make_user, as_user):
+    # Renaming, keeping or moving touches the row; the tray would announce the file as ready again.
+    from sqlalchemy import update
+
+    user = make_user()
+    old = _file(db, user.id, title="Old")
+    db.execute(
+        update(DownloadJob)
+        .where(DownloadJob.id == old.id)
+        .values(updated_at=NOW - timedelta(days=30), ready_at=NOW - timedelta(days=30))
+    )
+    db.commit()
+    c = as_user(user)
+    assert (
+        c.patch(f"/api/files/{old.id}", json={"name": "Renamed", "kept": True}).status_code == 200
+    )
+    assert [row["id"] for row in c.get("/api/downloads").json()] == []

backend/tests/test_migration_0102.py

@@ -71,3 +71,71 @@ def test_existing_jobs_are_placed_and_keep_their_expiry(db, make_user):
     }
     # alice: youtube, tiktok, upload, plex; bob: youtube.
     assert folders == 5
+
+
+def test_the_admins_retention_is_kept_and_the_downgrade_gives_the_asset_back_its_expiry(
+    db, make_user
+):
+    alice, bob = make_user().id, make_user().id
+    cfg = Config("alembic.ini")
+    command.downgrade(cfg, "0101_plex_text_plain")
+    try:
+        with engine.begin() as conn:
+            conn.execute(
+                text(
+                    "INSERT INTO system_config (key, value) VALUES ('download_retention_days', '7')"
+                    " ON CONFLICT (key) DO UPDATE SET value = '7'"
+                )
+            )
+            yt = conn.execute(
+                text(
+                    "INSERT INTO media_assets (source_kind, source_ref, format_sig, status,"
+                    " ref_count, expires_at) VALUES ('youtube', 'abcdefghijk', 'sig', 'ready', 2,"
+                    " '2030-01-02T00:00:00+00:00') RETURNING id"
+                )
+            ).scalar_one()
+            for user in (alice, bob):
+                conn.execute(
+                    text(
+                        "INSERT INTO download_jobs (user_id, asset_id, source_kind, source_ref,"
+                        " profile_snapshot, status) VALUES (:u, :a, 'youtube', 'abcdefghijk',"
+                        " '{}', 'done')"
+                    ),
+                    {"u": user, "a": yt},
+                )
+        command.upgrade(cfg, "head")
+        with engine.connect() as conn:
+            retention = conn.execute(
+                text(
+                    "SELECT DISTINCT retention_days FROM file_folders WHERE source_key = 'youtube'"
+                )
+            ).all()
+
+        def asset_expiry_after_downgrade(alice_at, bob_at):
+            with engine.begin() as conn:
+                for user, at in ((alice, alice_at), (bob, bob_at)):
+                    conn.execute(
+                        text(
+                            "UPDATE download_jobs SET expires_at = CAST(:at AS timestamptz)"
+                            " WHERE asset_id = :a AND user_id = :u"
+                        ),
+                        {"at": at, "a": yt, "u": user},
+                    )
+            command.downgrade(cfg, "0101_plex_text_plain")
+            with engine.connect() as conn:
+                got = conn.execute(
+                    text("SELECT expires_at::date FROM media_assets WHERE id = :a"), {"a": yt}
+                ).scalar()
+            command.upgrade(cfg, "head")
+            return str(got) if got else None
+
+        # The shared file goes when its last holder's copy would; one holder keeping it keeps it.
+        latest = asset_expiry_after_downgrade(
+            "2030-02-01T00:00:00+00:00", "2030-03-01T00:00:00+00:00"
+        )
+        kept_by_one = asset_expiry_after_downgrade("2030-02-01T00:00:00+00:00", None)
+    finally:
+        command.upgrade(cfg, "head")
+    assert retention == [(7,)]
+    assert latest == "2030-03-01"
+    assert kept_by_one is None

backend/tests/test_migration_0104.py

@@ -1,5 +1,5 @@
-"""Migration 0104's data step, DB-lane: a done file from before is dated by when it finished where
-its "ready to watch" notice says so, else by when it was asked for; a job not done has no date."""
+"""Migration 0104's data step, DB-lane: a done file from before is dated by when it was asked for
+(where it already sat); a job not done has no date."""
 
 from alembic import command
 from alembic.config import Config
@@ -8,7 +8,7 @@ from sqlalchemy import text
 from app.db import engine
 
 
-def test_a_done_file_takes_its_notice_time_else_its_request_time(db, make_user):
+def test_a_done_file_takes_its_request_time(db, make_user):
     uid = make_user().id
     cfg = Config("alembic.ini")
     command.downgrade(cfg, "0103_files_sha256")
@@ -25,17 +25,8 @@ def test_a_done_file_takes_its_notice_time_else_its_request_time(db, make_user):
                     {"u": uid, "s": status, "c": created},
                 ).scalar_one()
 
-            retried = job("done", "2026-10-01T10:00:00+00:00")
-            quiet = job("done", "2026-10-02T10:00:00+00:00")
+            done = job("done", "2026-10-02T10:00:00+00:00")
             failed = job("error", "2026-10-03T10:00:00+00:00")
-            for at in ("2026-10-05T09:00:00+00:00", "2026-10-06T09:00:00+00:00"):
-                conn.execute(
-                    text(
-                        "INSERT INTO notifications (user_id, type, title, data, created_at) VALUES"
-                        " (:u, 'download_done', 't', CAST(:d AS json), CAST(:c AS timestamptz))"
-                    ),
-                    {"u": uid, "d": f'{{"job_id": {retried}}}', "c": at},
-                )
     finally:
         command.upgrade(cfg, "head")
 
@@ -47,6 +38,5 @@ def test_a_done_file_takes_its_notice_time_else_its_request_time(db, make_user):
                 )
             ).all()
         )
-    assert rows[retried] == "10-06 09"  # its last finish
-    assert rows[quiet] == "10-02 10"  # no notice left: where it already sat
+    assert rows[done] == "10-02 10"
     assert rows[failed] is None

backend/tests/test_migration_0105.py

@@ -0,0 +1,45 @@
+"""Migration 0105's data step, DB-lane: a file in a source folder records that folder's key as its
+origin; a file in an own folder is left to be worked out, as before."""
+
+from alembic import command
+from alembic.config import Config
+from sqlalchemy import text
+
+from app.db import engine
+
+
+def test_a_file_in_a_source_folder_records_its_origin(db, make_user):
+    uid = make_user().id
+    cfg = Config("alembic.ini")
+    command.downgrade(cfg, "0104_job_ready_at")
+    try:
+        with engine.begin() as conn:
+
+            def folder(kind, key, name):
+                return conn.execute(
+                    text(
+                        "INSERT INTO file_folders (user_id, kind, source_key, name) VALUES"
+                        " (:u, :k, :s, :n) RETURNING id"
+                    ),
+                    {"u": uid, "k": kind, "s": key, "n": name},
+                ).scalar_one()
+
+            def job(folder_id):
+                return conn.execute(
+                    text(
+                        "INSERT INTO download_jobs (user_id, source_kind, source_ref,"
+                        " profile_snapshot, status, folder_id) VALUES (:u, 'url', 'x', '{}',"
+                        " 'queued', :f) RETURNING id"
+                    ),
+                    {"u": uid, "f": folder_id},
+                ).scalar_one()
+
+            in_source = job(folder("source", "site:vimeo.com", None))
+            in_own = job(folder("own", None, "Mine"))
+    finally:
+        command.upgrade(cfg, "head")
+
+    with engine.connect() as conn:
+        rows = dict(conn.execute(text("SELECT id, source_key FROM download_jobs")).all())
+    assert rows[in_source] == "site:vimeo.com"
+    assert rows[in_own] is None

backend/tests/test_storage.py

@@ -6,6 +6,7 @@ from datetime import date
 from app.downloads.storage import (
     MediaMeta,
     download_filename,
+    files_root_problem,
     rel_path,
     safe_abs_path,
     sanitize,
@@ -174,3 +175,24 @@ def test_sidecars_keep_the_full_basename_when_the_title_has_a_dot(tmp_path: Path
     assert f"{stem}.nfo" in names, names
     assert f"{stem}.jpg" in names, names
     assert "Chan_-_2026-08-12_-_Sonka_Hazilag.nfo" not in names  # the old truncated orphan
+
+
+def test_a_files_root_on_the_containers_own_disk_is_refused(tmp_path):
+    # The image's empty /files, left unmounted by an old compose file: every file would 404 and
+    # every download would go with the container.
+    assert "not a mounted volume" in files_root_problem(
+        str(tmp_path), system_root=str(tmp_path.parent)
+    )
+    assert "does not exist" in files_root_problem(str(tmp_path / "gone"))
+    # On another filesystem than the container's own, and writable: the volume.
+    assert files_root_problem(str(tmp_path), system_root="/proc") is None
+
+
+def test_the_old_name_of_the_files_root_still_finds_the_files(monkeypatch):
+    from app.config import Settings
+
+    monkeypatch.delenv("FILES_ROOT", raising=False)
+    monkeypatch.setenv("DOWNLOAD_ROOT", "/downloads")
+    assert Settings().files_root == "/downloads"
+    monkeypatch.setenv("FILES_ROOT", "/files")
+    assert Settings().files_root == "/files"

backend/tests/test_worker_knobs.py

@@ -67,6 +67,7 @@ def test_worker_boots_with_the_stored_concurrency_and_log_level(monkeypatch, sto
     monkeypatch.setattr(worker, "_wait_for_schema", lambda: True)
     monkeypatch.setattr(worker.logsink, "start", lambda: None)
     monkeypatch.setattr(worker, "_recover_orphans", lambda: None)
+    monkeypatch.setattr(worker.storage, "files_root_problem", lambda _root: None)
 
     threading.Timer(0.5, worker._stop.set).start()
     worker.main()
@@ -107,3 +108,17 @@ def test_an_unreadable_database_keeps_what_runs(monkeypatch, levels, loops):
     assert worker._target == 2
     assert sorted(threads) == [0, 1]  # the loops still run on the boot value
     assert levels == []
+
+
+def test_the_worker_will_not_start_on_a_files_root_that_is_not_the_volume(monkeypatch, loops):
+    # Files downloaded into the container's own empty /files go with the container: refuse, loudly.
+    monkeypatch.setattr(worker.settings, "worker_enabled", True)
+    monkeypatch.setattr(worker.logs, "configure_logging", lambda _source: None)
+    monkeypatch.setattr(worker.signal, "signal", lambda *_a: None)
+    monkeypatch.setattr(worker, "_wait_for_schema", lambda: True)
+    monkeypatch.setattr(worker.logsink, "start", lambda: None)
+    monkeypatch.setattr(worker.storage, "files_root_problem", lambda _root: "not a mounted volume")
+    with pytest.raises(SystemExit) as exit_:
+        worker.main()
+    assert exit_.value.code == 1
+    assert loops == []