soulsync/core/library/embedded_id_reconcile.py
BoulderBadgeDad 83c1cd92aa Auto-reconcile embedded IDs for new tracks on library scans
Extends the manual "Import IDs from File Tags" backfill so newly-scanned
files get their embedded provider IDs pulled into the DB automatically —
no button press needed to keep up with new music.

How it works:
- insert_or_update_media_track now returns 'inserted' / 'updated' / False
  (truthy-compatible; existing `if track_success` callers unaffected) so
  the scan worker can tell a genuinely new row from an update.
- DatabaseUpdateWorker collects the ids it newly INSERTED this run
  (self._new_track_ids) across all insert paths (Plex/Jellyfin/deep).
- After run()/run_deep_scan(), web_server calls _reconcile_after_scan(),
  which gap-fills embedded IDs for just those new tracks. Runs as a
  post-scan pass (the scan loop itself is untouched/fast — the media
  server API never exposes these custom IDs, so the file must be read
  once regardless; batching at the end keeps it out of the hot loop and
  best-effort so it can never abort a scan). A progress phase ("Reading
  file tags for N new tracks…") surfaces the full-refresh tail.

Shared engine:
- New reconcile_library() in core does the paging + lazy parent-map
  loading (only loads albums/artists actually referenced — cheap when
  scoped to a few new tracks) + per-page commits. BOTH the manual button
  and the scan hook call it, so there's one tested orchestration, no
  duplication. The backfill job was refactored onto it.

Same hardened safety: gap-fill only, atomically guarded against
overwrite, schema-introspected, idempotent. Scoped to new arrivals for
incremental/deep; full refresh re-inserts everything as new (recovering
the IDs a full-refresh wipe destroys).

+10 reconcile tests (reconcile_library scope/idempotency/progress/stop +
the engine). Full suite clean (only pre-existing soundcloud /app env
failures remain).
2026-06-05 18:31:11 -07:00

415 lines
17 KiB
Python

"""Reconcile provider IDs embedded in audio files into the library DB.
Enrichment workers (Spotify / iTunes / MusicBrainz / Deezer / Tidal /
AudioDB / Genius / Last.fm) resolve each artist / album / track to a provider ID
via API calls, gating their work queues on ``{provider}_match_status IS
NULL``. But files that SoulSync (or MusicBrainz Picard) already tagged
carry those IDs in their metadata. Reading them back and gap-filling the
``{provider}_id`` + ``{provider}_match_status = 'matched'`` columns lets
the workers skip the API lookup entirely — large API savings on an
already-tagged library.
Split into a PURE planning layer and a thin DB apply layer:
- :func:`plan_reconcile` takes the tags read from ONE file (via
``core.library.file_tags.read_embedded_tags``) plus the current IDs of
that file's track + its parent album + artist, and produces the list of
:class:`Fill` operations to perform. It is gap-fill only: a provider id
that already has a value is never planned for change; a DISAGREEING
embedded id is reported as a conflict instead.
- :func:`apply_reconcile_plan` writes a plan, one guarded ``UPDATE`` per
id column: ``WHERE id = ? AND ({id_col} IS NULL OR {id_col} = '')``.
The guard makes the gap-fill ATOMIC — even if an enrichment worker
matched the same entity between the plan's read and this write, the
fill simply affects 0 rows instead of clobbering the worker's value.
Columns are introspected first so a schema version missing a provider's
columns is skipped, not errored.
Scope note: the MusicBrainz *recording* (track) ID is intentionally not
reconciled — on ID3 it lives in a ``UFID`` frame the shared reader
doesn't surface and the Vorbis ``musicbrainz_trackid`` convention is
format-ambiguous. MB *album* and *artist* IDs (which drive most worker
API calls) ARE reconciled, as are the clean per-provider track/album/
artist IDs of the other services.
"""
from __future__ import annotations
from dataclasses import dataclass, field
from typing import Any, Dict, List, Optional
# Each entry: (embedded-tag key from read_embedded_tags, entity, id column,
# match-status column). The id columns mirror web_server._SERVICE_ID_COLUMNS;
# they're spelled out here so this module stays importable without the Flask
# app. Single-column providers (deezer/tidal/audiodb/genius) reuse one id
# column across entity types — that's fine, fills are keyed by (entity, col).
_RECONCILE_FIELDS = (
('spotify_track_id', 'track', 'spotify_track_id', 'spotify_match_status'),
('spotify_album_id', 'album', 'spotify_album_id', 'spotify_match_status'),
('spotify_artist_id', 'artist', 'spotify_artist_id', 'spotify_match_status'),
('itunes_track_id', 'track', 'itunes_track_id', 'itunes_match_status'),
('itunes_album_id', 'album', 'itunes_album_id', 'itunes_match_status'),
('itunes_artist_id', 'artist', 'itunes_artist_id', 'itunes_match_status'),
('musicbrainz_albumid', 'album', 'musicbrainz_release_id', 'musicbrainz_match_status'),
('musicbrainz_artistid', 'artist', 'musicbrainz_id', 'musicbrainz_match_status'),
('deezer_track_id', 'track', 'deezer_id', 'deezer_match_status'),
('deezer_album_id', 'album', 'deezer_id', 'deezer_match_status'),
('deezer_artist_id', 'artist', 'deezer_id', 'deezer_match_status'),
('tidal_track_id', 'track', 'tidal_id', 'tidal_match_status'),
('tidal_album_id', 'album', 'tidal_id', 'tidal_match_status'),
('tidal_artist_id', 'artist', 'tidal_id', 'tidal_match_status'),
('audiodb_track_id', 'track', 'audiodb_id', 'audiodb_match_status'),
('audiodb_album_id', 'album', 'audiodb_id', 'audiodb_match_status'),
('audiodb_artist_id', 'artist', 'audiodb_id', 'audiodb_match_status'),
('genius_track_id', 'track', 'genius_id', 'genius_match_status'),
# Last.fm embeds a single LASTFM_URL — sourced from get_track_info(), so it
# is the TRACK's url. Map to tracks.lastfm_url only (artist/album last.fm
# urls are different urls and aren't carried in the file).
('lastfm_url', 'track', 'lastfm_url', 'lastfm_match_status'),
)
_ENTITIES = ('track', 'album', 'artist')
_ENTITY_TABLE = {'track': 'tracks', 'album': 'albums', 'artist': 'artists'}
@dataclass(frozen=True)
class Fill:
"""One provider-id column to gap-fill on one entity."""
entity: str # 'track' | 'album' | 'artist'
id_column: str # e.g. 'spotify_artist_id'
status_column: str # e.g. 'spotify_match_status'
value: str # the embedded id to write
@dataclass
class ReconcilePlan:
"""The outcome of planning one file against its current DB rows.
``fills`` are the gap-fill operations to apply (empty id columns only).
``already_present`` counts embedded ids that matched a value already
stored (no-op). ``conflicts`` lists embedded ids that DISAGREE with a
stored value — never applied, surfaced for review.
"""
fills: List[Fill] = field(default_factory=list)
already_present: int = 0
conflicts: List[Dict[str, str]] = field(default_factory=list)
@property
def filled(self) -> int:
return len(self.fills)
@property
def has_updates(self) -> bool:
return bool(self.fills)
def fills_for(self, entity: str) -> List[Fill]:
return [f for f in self.fills if f.entity == entity]
@dataclass
class ReconcileApplied:
"""Counts from actually writing a plan (based on real ``rowcount``)."""
rows_updated: int = 0 # distinct entity rows touched
ids_filled: int = 0 # id columns that actually landed (guard passed)
def _clean(value: Any) -> Optional[str]:
"""Normalise a tag/column value to a non-empty stripped string or None."""
if value is None:
return None
s = str(value).strip()
return s or None
def plan_reconcile(
embedded_tags: Optional[Dict[str, Any]],
current_ids: Optional[Dict[str, Dict[str, Any]]],
) -> ReconcilePlan:
"""Plan which provider-ID columns to gap-fill from one file's tags.
Args:
embedded_tags: the ``tags`` dict from ``read_embedded_tags`` (flat
``friendly_key -> value``). ``None`` / empty yields an empty plan.
current_ids: ``{'track': {...}, 'album': {...}, 'artist': {...}}``
where each inner dict holds the entity's CURRENT column values
(at minimum the id columns this module touches). Missing
entities / keys are treated as empty (eligible to fill).
Returns:
A :class:`ReconcilePlan`. Gap-fill only — an id column with any
existing value is never planned; a disagreeing embedded id is
recorded in ``conflicts``.
"""
plan = ReconcilePlan()
tags = embedded_tags or {}
current = current_ids or {}
queued: Dict[tuple, str] = {} # (entity, id_col) already queued this pass
for embedded_key, entity, id_col, status_col in _RECONCILE_FIELDS:
new_val = _clean(tags.get(embedded_key))
if not new_val:
continue
row = current.get(entity) or {}
existing = _clean(row.get(id_col))
if existing is not None:
if existing != new_val:
plan.conflicts.append({
'entity': entity, 'column': id_col,
'existing': existing, 'embedded': new_val,
})
else:
plan.already_present += 1
continue
key = (entity, id_col)
if key in queued:
# A single-column provider already queued this id col this pass.
if queued[key] != new_val:
plan.conflicts.append({
'entity': entity, 'column': id_col,
'existing': queued[key], 'embedded': new_val,
})
continue
queued[key] = new_val
plan.fills.append(Fill(entity, id_col, status_col, new_val))
return plan
@dataclass
class TrackReconcileResult:
"""Outcome of reconciling one track row against its file's tags."""
applied: 'ReconcileApplied'
conflicts: int = 0
readable: bool = True # False when the file's tags couldn't be read
def reconcile_track_row(
cursor,
track_row: Dict[str, Any],
album_map: Dict[str, Dict[str, Any]],
artist_map: Dict[str, Dict[str, Any]],
embedded_tags: Optional[Dict[str, Any]],
) -> TrackReconcileResult:
"""Reconcile one track row + its parent album/artist against one file.
Pure orchestration over :func:`plan_reconcile` / :func:`apply_reconcile_plan`,
extracted so the per-track logic (id extraction, plan→apply chaining,
keeping the in-memory parent maps fresh for sibling tracks) is testable
without the Flask job. ``embedded_tags`` is the ``tags`` dict from
``read_embedded_tags`` (``None`` => unreadable file).
``album_map`` / ``artist_map`` map entity-id -> current column dict; this
function UPDATES them in place with any fills it applies so a later track
on the same album/artist sees the value and doesn't re-plan it. (DB safety
is the guarded UPDATE in apply, never these maps.)
"""
if not embedded_tags:
return TrackReconcileResult(ReconcileApplied(), 0, readable=False)
album_id = str(track_row['album_id']) if track_row.get('album_id') is not None else None
artist_id = str(track_row['artist_id']) if track_row.get('artist_id') is not None else None
plan = plan_reconcile(embedded_tags, {
'track': track_row,
'album': album_map.get(album_id, {}) if album_id else {},
'artist': artist_map.get(artist_id, {}) if artist_id else {},
})
applied = apply_reconcile_plan(cursor, {
'track': track_row.get('id'), 'album': album_id, 'artist': artist_id,
}, plan)
if album_id:
for f in plan.fills_for('album'):
album_map.setdefault(album_id, {})[f.id_column] = f.value
if artist_id:
for f in plan.fills_for('artist'):
artist_map.setdefault(artist_id, {})[f.id_column] = f.value
return TrackReconcileResult(applied, len(plan.conflicts), readable=True)
@dataclass
class ReconcileTotals:
"""Accumulated counts over a reconcile run."""
total: int = 0
processed: int = 0
ids_filled: int = 0
entities_updated: int = 0
conflicts: int = 0
unreadable: int = 0
def _load_missing_rows(cursor, ids, table, target: Dict[str, Dict[str, Any]]) -> None:
"""Load any not-yet-cached entity rows for ``ids`` into ``target`` in place.
Ids with no row get an empty dict so they're never re-queried. Chunked to
keep the IN clause bounded.
"""
missing = [i for i in {x for x in ids if x} if i not in target]
for start in range(0, len(missing), 500):
chunk = missing[start:start + 500]
ph = ','.join('?' * len(chunk))
cursor.execute(f"SELECT * FROM {table} WHERE id IN ({ph})", chunk)
for r in cursor.fetchall():
target[str(r['id'])] = dict(r)
for i in missing:
target.setdefault(i, {}) # mark absent → don't re-query
def reconcile_library(
conn,
read_tags,
track_ids=None,
page_size: int = 500,
on_progress=None,
should_stop=None,
) -> ReconcileTotals:
"""Gap-fill embedded provider IDs into the DB for a set of tracks.
Shared orchestration used by both the manual backfill job and the
auto-reconcile hook on library scans. Pages the track list (bounded
memory), lazily loads only the parent album/artist rows actually
referenced (cheap when scoped to a handful of new tracks), and commits
per page so concurrent enrichment workers aren't starved of the write
lock.
Args:
conn: open DB connection; this function commits per page.
read_tags: callable ``(file_path) -> tags dict | None``. The caller
injects path resolution + ``read_embedded_tags`` so this module
stays free of Flask / docker-path concerns. ``None`` => unreadable.
track_ids: iterable of track ids to reconcile, or ``None`` for every
track that has a ``file_path``.
page_size: rows materialised per page.
on_progress: optional ``(totals, current_title) -> None`` after each
track (for live UI).
should_stop: optional ``() -> bool`` checked between tracks/pages to
abort early.
Returns:
:class:`ReconcileTotals`.
"""
from utils.logging_config import get_logger
logger = get_logger("library.reconcile")
totals = ReconcileTotals()
cur = conn.cursor()
if track_ids is None:
cur.execute("SELECT id FROM tracks WHERE file_path IS NOT NULL AND TRIM(file_path) != ''")
ids = [str(r[0]) for r in cur.fetchall()]
else:
ids = [str(t) for t in track_ids if t is not None]
totals.total = len(ids)
album_map: Dict[str, Dict[str, Any]] = {}
artist_map: Dict[str, Dict[str, Any]] = {}
for start in range(0, len(ids), page_size):
if should_stop and should_stop():
break
page = ids[start:start + page_size]
ph = ','.join('?' * len(page))
cur.execute(f"SELECT * FROM tracks WHERE id IN ({ph})", page)
rows = [dict(r) for r in cur.fetchall()]
_load_missing_rows(cur, [str(r['album_id']) for r in rows if r.get('album_id') is not None],
'albums', album_map)
_load_missing_rows(cur, [str(r['artist_id']) for r in rows if r.get('artist_id') is not None],
'artists', artist_map)
for tr in rows:
if should_stop and should_stop():
break
title = tr.get('title') or '?'
try:
tags = read_tags(tr.get('file_path'))
result = reconcile_track_row(cur, tr, album_map, artist_map, tags)
if not result.readable:
totals.unreadable += 1
else:
totals.ids_filled += result.applied.ids_filled
totals.entities_updated += result.applied.rows_updated
totals.conflicts += result.conflicts
except Exception as e:
logger.debug("reconcile: skipped track %s: %s", tr.get('id'), e)
totals.unreadable += 1
finally:
totals.processed += 1
if on_progress:
on_progress(totals, title)
conn.commit()
return totals
def _existing_columns(cursor, table: str) -> set:
"""Return the set of column names on ``table`` (migration-safe guard)."""
cursor.execute(f"PRAGMA table_info({table})")
return {r[1] for r in cursor.fetchall()}
def apply_reconcile_plan(cursor, entity_ids: Dict[str, Any], plan: ReconcilePlan) -> ReconcileApplied:
"""Apply a :class:`ReconcilePlan` to the DB via ``cursor``.
Each fill is a single guarded ``UPDATE``:
UPDATE {table} SET {id}=?, {status}='matched', {attempted}=now
WHERE id=? AND ({id} IS NULL OR {id}='')
The ``id IS NULL OR id=''`` guard makes the gap-fill atomic: if the
column became non-empty between the plan's read and now (an enrichment
worker matched it concurrently), the UPDATE affects 0 rows and the
worker's value is preserved. Only columns that exist on the table are
written (introspected + cached per call), so a schema missing a
provider's columns is silently skipped.
Args:
cursor: an open DB cursor (caller owns the transaction/commit).
entity_ids: ``{'track': id, 'album': id, 'artist': id}``. An entity
with no id is skipped.
Returns:
A :class:`ReconcileApplied` with counts derived from real rowcounts.
"""
result = ReconcileApplied()
touched: set = set()
col_cache: Dict[str, set] = {}
for fill in plan.fills:
ent_id = entity_ids.get(fill.entity)
if ent_id is None or ent_id == '':
continue
table = _ENTITY_TABLE[fill.entity]
if table not in col_cache:
col_cache[table] = _existing_columns(cursor, table)
cols = col_cache[table]
if fill.id_column not in cols:
continue
assignments = [f"{fill.id_column} = ?"]
values: List[Any] = [fill.value]
if fill.status_column in cols:
assignments.append(f"{fill.status_column} = ?")
values.append('matched')
attempted = fill.status_column.replace('_match_status', '_last_attempted')
if attempted in cols:
assignments.append(f"{attempted} = CURRENT_TIMESTAMP")
cursor.execute(
f"UPDATE {table} SET {', '.join(assignments)} "
f"WHERE id = ? AND ({fill.id_column} IS NULL OR {fill.id_column} = '')",
values + [str(ent_id)],
)
if cursor.rowcount:
result.ids_filled += 1
touched.add((fill.entity, str(ent_id)))
result.rows_updated = len(touched)
return result