feat: images and attachments go to the off-site bucket once each, by fingerprint

With an off-site destination set up, the library's files are sent to the
bucket's media/ folder as media/ab/cd/<sha256><ext>, beside the database
dumps under database/. Each file goes once: a new backup_blob table records
what each destination holds (migration 0121), and a run's work list is every
image and attachment sha with no row yet, so runs are incremental and a new
bucket starts its copy from the beginning.

- The bucket checks each file against the sha recorded at import; a file
  over 1 GiB goes in parts and is checked before the upload is completed.
  A file missing or changed on disk is reported, not stored.
- A run is chunked: each 45-minute chunk commits after every file and queues
  the next, so other long-maintenance jobs get turns. One run at a time.
  The stall sweep judges a media run by its progress heartbeat
  (backup_run.last_progress_at), not its start.
- A file that can't be sent is listed on the run (status ok, with the error
  line) and tried again by the next run. A refused put for an object
  already in an append-only bucket counts as sent.
- Nightly, with a destination set, runs the dump and then the media.
- Weekly reconcile lists media/ and re-sends what the bucket lost, and
  adopts what it already holds (a restored database sends nothing twice).
- Uploads use the bandwidth limit; read transactions end before transfers.
- Backups card: with a destination, an "Images and attachments" section
  shows coverage (files and bytes in the bucket), "Send new files now", and
  the media runs. Without one, the local tarball section is unchanged.
- Media run history keeps the latest 60 rows.

Task #6879, milestone 586.

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

backend/app/tasks/backup.py

@@ -8,6 +8,7 @@ BackupRun.
 from __future__ import annotations
 
 import logging
+import time
 from datetime import UTC, datetime
 from pathlib import Path
 
@@ -18,7 +19,7 @@ from sqlalchemy.exc import DBAPIError, OperationalError
 from ..celery_app import celery
 from ..config import get_config
 from ..models import BackupRun, ImportSettings
-from ..services import backup_destination, backup_service
+from ..services import backup_destination, backup_media, backup_s3, backup_service
 from ._sync_engine import sync_session_factory as _sync_session_factory
 
 log = logging.getLogger(__name__)
@@ -267,7 +268,9 @@ def restore_images_task(self, *, source_backup_run_id: int) -> dict:
 def prune_backups() -> dict:
     """Daily Beat. Per-kind retention from ImportSettings.
 
-    Returns {"db_deleted": N, "images_deleted": M, "files_unlinked": K}.
+    Returns {"db_deleted": N, "images_deleted": M, "media_deleted": L,
+    "files_unlinked": K}. Media runs leave no files here; only their history
+    is trimmed, to the latest MEDIA_RUNS_KEPT.
     Tagged rows (tag IS NOT NULL) are never pruned.
     Status='running' / 'restoring' rows are never pruned — the
     recover_stalled_backup_runs sweep flips truly-stuck ones to
@@ -276,12 +279,13 @@ def prune_backups() -> dict:
     Audit 2026-06-02 added the dedicated sweep.)
     """
     SessionLocal = _sync_session_factory()
-    counts = {"db_deleted": 0, "images_deleted": 0, "files_unlinked": 0}
+    counts = {"db_deleted": 0, "images_deleted": 0, "media_deleted": 0, "files_unlinked": 0}
     with SessionLocal() as session:
         s = ImportSettings.load_sync(session)
         for kind, keep in (
             ("db", s.backup_db_keep_last_n),
             ("images", s.backup_images_keep_last_n),
+            ("media", MEDIA_RUNS_KEPT),
         ):
             candidates = session.execute(
                 select(BackupRun)
@@ -325,4 +329,168 @@ def backup_db_nightly() -> dict:
     if now_hour != configured_hour:
         return {"skipped": f"hour={now_hour} != configured={configured_hour}"}
     res = backup_db_task.delay(triggered_by="nightly")
-    return {"dispatched": res.id}
+    out = {"dispatched": res.id}
+    # With an off-site destination, the night's backup is the whole library:
+    # the database dump, then any images and attachments not yet in the bucket.
+    # Both run on the long-maintenance lane, one after the other.
+    with SessionLocal() as session:
+        try:
+            offsite = backup_destination.saved_target(ImportSettings.load_sync(session))
+        except Exception:  # noqa: BLE001 — the media run reports it itself
+            offsite = True
+    if offsite is not None:
+        out["media_dispatched"] = backup_media_task.delay(triggered_by="nightly").id
+    return out
+
+
+# ── Off-site copy of the library's files ─────────────────────────────────────
+
+# Each task invocation sends files for this long, then queues the next chunk
+# of the same run, so other long-maintenance jobs get their turn in between.
+MEDIA_CHUNK_SECONDS = 45 * 60
+MEDIA_RUNS_KEPT = 60
+# A failed file is listed on the run, up to this many; the count covers all.
+MEDIA_FAILURES_LISTED = 20
+
+
+def _offsite_manifest(target) -> dict:
+    return {"endpoint": target.endpoint, "bucket": target.bucket,
+            "key": backup_media.PREFIX}
+
+
+@celery.task(
+    name="backend.app.tasks.backup.backup_media_task",
+    bind=True,
+    autoretry_for=(OperationalError, DBAPIError),
+    retry_backoff=30, retry_backoff_max=300, max_retries=2,
+    # A chunk stops starting files after MEDIA_CHUNK_SECONDS; the limits leave
+    # room for the file in flight, a multi-gigabyte archive at a low upload
+    # limit included.
+    soft_time_limit=21600, time_limit=23400,
+)
+def backup_media_task(self, *, run_id: int | None = None,
+                      triggered_by: str = "manual") -> dict:
+    """Send the library's images and attachments that aren't in the off-site
+    bucket yet, one chunk of a run. Called without `run_id` it starts a run
+    (unless one is already going); each chunk queues the next until nothing is
+    left. A file that can't be sent is listed on the run and tried again by
+    the next run. Returns {'backup_run_id': N, ...}."""
+    SessionLocal = _sync_session_factory()
+    now = datetime.now(UTC)
+    with SessionLocal() as session:
+        if run_id is None:
+            going = session.execute(
+                select(BackupRun.id).where(BackupRun.kind == "media")
+                .where(BackupRun.status == "running").limit(1)
+            ).scalar_one_or_none()
+            if going is not None:
+                return {"backup_run_id": going, "skipped": "a media backup is already running"}
+            row = BackupRun(kind="media", status="running", triggered_by=triggered_by,
+                            started_at=now, last_progress_at=now, size_bytes=0,
+                            manifest={"sent": 0, "failed": 0, "failures": [], "cursor": ""})
+            session.add(row)
+            session.commit()
+            run_id = row.id
+        else:
+            row = session.get(BackupRun, run_id)
+            if row is None or row.status != "running":
+                return {"backup_run_id": run_id, "skipped": "run is no longer running"}
+
+    try:
+        with SessionLocal() as session:
+            settings = ImportSettings.load_sync(session)
+            target = backup_destination.saved_target(settings)
+            limit_mbps = settings.backup_upload_limit_mbps
+        if target is None:
+            raise backup_s3.S3Error("No off-site destination is set up.")
+        more = _send_media_chunk(SessionLocal, run_id, target, backup_s3.Pacer(limit_mbps))
+    except (SoftTimeLimitExceeded, Exception) as exc:
+        with SessionLocal() as session:
+            row = session.get(BackupRun, run_id)
+            if row is not None:
+                _mark_failed(session, row, exc)
+        raise
+
+    if more:
+        backup_media_task.delay(run_id=run_id)
+        return {"backup_run_id": run_id, "continues": True}
+    with SessionLocal() as session:
+        row = session.get(BackupRun, run_id)
+        m = dict(row.manifest or {})
+        row.status = "ok"
+        row.finished_at = datetime.now(UTC)
+        if m.get("failed"):
+            first = (m.get("failures") or [{}])[0]
+            row.error = (f"{m['failed']} file(s) weren't sent; the next run tries them again. "
+                         f"First: {first.get('path')}: {first.get('error')}")[:2000]
+        session.commit()
+    return {"backup_run_id": run_id, "continues": False}
+
+
+def _send_media_chunk(SessionLocal, run_id: int, target, pacer) -> bool:
+    """Send files until the chunk's time is up or none are left. Progress is
+    committed after every file, so a chunk that dies loses at most the file
+    in flight. Returns True when files remain for another chunk."""
+    tid = backup_media.target_id(target)
+    deadline = time.monotonic() + MEDIA_CHUNK_SECONDS
+    with SessionLocal() as session:
+        row = session.get(BackupRun, run_id)
+        m = dict(row.manifest or {})
+        m["offsite"] = _offsite_manifest(target)
+        cursor = m.get("cursor") or ""
+        while True:
+            batch = backup_media.pending(session, tid, after=cursor)
+            # End the read before any upload starts: a transaction left open
+            # across a long transfer holds back vacuum for its whole length.
+            session.commit()
+            if not batch:
+                m["cursor"] = cursor
+                row.manifest = dict(m)
+                row.last_progress_at = datetime.now(UTC)
+                session.commit()
+                return False
+            for blob in batch:
+                if time.monotonic() > deadline:
+                    m["cursor"] = cursor
+                    row.manifest = dict(m)
+                    session.commit()
+                    return True
+                try:
+                    key = backup_media.send(target, blob, IMAGES_ROOT, pacer)
+                except (backup_s3.S3Error, OSError) as exc:
+                    m["failed"] = m.get("failed", 0) + 1
+                    if len(m.get("failures") or []) < MEDIA_FAILURES_LISTED:
+                        m["failures"] = [*(m.get("failures") or []),
+                                         {"path": blob.path, "error": str(exc)[:300]}]
+                else:
+                    backup_media.record(session, tid, [(blob, key)])
+                    m["sent"] = m.get("sent", 0) + 1
+                    row.size_bytes = (row.size_bytes or 0) + blob.size_bytes
+                cursor = blob.sha256
+                m["cursor"] = cursor
+                row.manifest = dict(m)
+                row.last_progress_at = datetime.now(UTC)
+                session.commit()
+
+
+@celery.task(
+    name="backend.app.tasks.backup.reconcile_media_backup",
+    soft_time_limit=1800, time_limit=2100,
+)
+def reconcile_media_backup() -> dict:
+    """Weekly Beat. Compare what the off-site bucket holds under `media/` with
+    what FabledCurator believes it sent: a missing object is sent again by
+    the next run, and an object already there is not sent twice. Skipped
+    while a media run is going, and when no destination is set up."""
+    SessionLocal = _sync_session_factory()
+    with SessionLocal() as session:
+        target = backup_destination.saved_target(ImportSettings.load_sync(session))
+        if target is None:
+            return {"skipped": "no destination"}
+        going = session.execute(
+            select(BackupRun.id).where(BackupRun.kind == "media")
+            .where(BackupRun.status == "running").limit(1)
+        ).scalar_one_or_none()
+        if going is not None:
+            return {"skipped": "a media backup is running"}
+        return backup_media.reconcile(session, target)