Draining the queue via Soulseek was stalling: slow peers (~100KB/s) never finished a
full FLAC album within the fixed 10-min timeout, got cancelled at ~89%, and the
fall-through restarted from scratch on the next peer — while single-song folders named
like the album were also being tried.
- Stall detection: abandon a peer with no byte progress for ~90s (queued/dead) instead
of waiting the full timeout; a peer that IS progressing keeps its slot up to a 60-min
backstop, so a slow-but-working transfer can finish.
- Rank search candidates by peer: free upload slot, then speed, then shortest queue.
- Drop candidates with far fewer tracks than the release (single-song folders) up front,
so no download attempt is wasted on them.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
With qualityCutoff=3 (want hi-res), a Soulseek-grabbed CD copy stays below cutoff and
is re-attempted for upgrade. Without a guard, when Qobuz is gated by pacing those
attempts would (a) re-search every owned album each monitor cycle and (b) re-download
a same-quality Soulseek copy that keep-if-better discards — pure churn.
Add an upgrade guard: a >= CD-lossless copy can only be improved by hi-res Qobuz, so
when Qobuz is gated, skip the job WITHOUT searching (pre-search skip); and as a safety
net, skip the download when the best-ranked candidate can't beat the existing copy.
Force jobs always proceed. Extracted the shared keep-existing path into a helper.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two compounding causes of 'many downloads of one album from different sources' in
slskd: (1) the slskd client raised on timeout/error WITHOUT cancelling the transfer
it enqueued, so as the pipeline fell through to other peers each left a transfer
running; (2) the incomplete-download fall-through was uncapped, grinding through
every candidate of a popular album (~200 Soulseek peers), amplified now that Qobuz
pacing pushes most jobs to Soulseek.
- SlskdClient.download now cancels its enqueued transfers (DELETE) on any abandon.
- The download loop caps at _MAX_DOWNLOAD_ATTEMPTS (6) candidates.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Avoid re-triggering a USER_BLOCKED account by capping Qobuz to a human-looking
volume and cadence. A per-album gate (worker/lyra_worker/qobuz_gate.py) allows
Qobuz only within active hours, under today's cap (a warm-up ramp of a steady
cap), and past a randomized spacing since the last Qobuz download; the gap spreads
the day's budget across active hours with jitter. When the gate is closed the
pipeline drops Qobuz candidates so the job falls through to another source (and
monitored albums upgrade to Qobuz later); a Qobuz-only setup is never gated.
Config keys read live each loop; counters reset at the day boundary via the DB
clock. Tunable in Settings -> Qobuz -> Pacing (new /api/qobuz/pacing route).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Completeness/duration verification was applied only to the first candidate that
downloaded successfully; if that copy was incomplete or too short the whole job
failed, even when a later-ranked source had a complete copy (e.g. John Mayer 'The
Search for Everything' — Qobuz reliably fails one hi-res track, so it never fell
back to the deluxe edition or Soulseek).
Extract the checks into a pure _download_problem() and run it INSIDE the download
loop: a candidate that downloads but fails verification now falls through to the
next-ranked source, and the job only fails (needs_attention) when every candidate
is incomplete — reporting the last shortfall reason.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The completeness + duration guards used MusicBrainz's target.track_count /
total_duration_s, but MB picks an arbitrary releases[0] that is often a
deluxe/expanded edition with more tracks (and runtime) than the standard album a
source legitimately delivers. So complete standard-edition downloads were falsely
rejected — The Script 'No Sound Without Silence' (11 vs MB 12) as 'incomplete
download', Skillet 'Unleashed' (12 vs MB's 20-track deluxe) as 'too short'.
Now completeness is judged against the chosen source's own track count
(winner.track_count): a truncated download (fewer files than the source promised)
still fails, and the duration expectation is scaled to the delivered edition's
size. A separate 'implausibly small vs MB' guard (source has <= half MB's tracks)
still rejects a lone-track 'full album' video. Genuine partials (e.g. John Mayer,
a Qobuz track failing mid-download) still correctly fail.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A literal NUL had crept into the `${r.artist} ${r.album}` dedupe key on the
Floor, making git treat queue.tsx as a binary file (unreviewable diffs). Restore
the intended space; behavior is unchanged (the key stays unique per album).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The worker MB resolver picked release groups on (title_ratio, is_album)
and never scored the artist — even though resolve() knows the followed
artist. For a generic title, same-titled release groups by *different*
artists tie on title, and the is_album tiebreak (added in 06e7f91 for
Off the Wall) then actively preferred a foreign Album over the correct
release. Concretely, following Lake Street Dive's "Fun Machine" EP
resolved to an unrelated band "Bastards of Melody"'s same-titled Album,
which then propagated as the downloaded/imported artist.
Make credited-artist similarity the primary sort key, above title and
above is_album. It's a soft signal, never a hard filter, so credited-
name variations (feat., punctuation) still resolve; is_album now only
breaks ties within the same artist+title, preserving the Off the Wall
fix. Verified live: resolve("Lake Street Dive", "Fun Machine") now
returns the 6-track LSD EP.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Strengthen test to prove the job's paused state remains unmodified when
pause/unpause is rejected for a downloading job. Tests now seeds paused=true,
attempts paused=false, verifies 400, and fetches from DB to assert paused is
still true — catching regressions that would mutate before the guard check.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
MusicBrainz search returns release-groups tied on title similarity in an
arbitrary order; `max(groups, key=_ratio)` kept the first, so Michael
Jackson's "Off the Wall" resolved to the 2-track *single* release-group
(Off the Wall / Get on the Floor) instead of the 10-track album. Because
plan_track_names labels files purely by position, that short tracklist got
mapped onto the full album — track 2 became "Get on the Floor" instead of
"Rock with You", and the real "Off the Wall"/"Get on the Floor" files kept
their streamrip names, yielding duplicate titles.
Extract selection into a pure `_best_release_group()` that breaks title-
similarity ties toward primary-type == "Album". Similarity still dominates;
album preference only settles ties. Live-verified: Off the Wall now resolves
to the 10-track album with track 2 = "Rock With You".
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Last.fm's artist MBID can point at the wrong same-named band (it mapped
"Looking Glass" to an Australian stoner band, not the US pop group), and the
/lastfm rows trusted it verbatim — so opening/following grabbed the wrong
artist and the grabbed album never showed on the artist page. Now open/follow
resolve via MB search: resolveArtistChoice keeps candidates whose name matches
exactly — 1 match is used directly (even over a disagreeing Last.fm MBID), 2+
surface an ArtistPicker showing each MB disambiguation ("70's US pop group,
track 'Brandy'" vs "Australian stoner rock") with MB's top hit flagged
Suggested. Core decision logic unit-tested; no worker/schema change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A Qobuz login/metadata/artwork phase (and a Soulseek peer-queue or YouTube
info-fetch) can sit at 0% for a few minutes before byte progress flows, which
read as a stuck bar. While a downloading job's progress is still 0, render the
indeterminate "preparing…" bar (same style as the searching/finishing phases);
it flips to the determinate % bar on the first reported byte and never back
(the progress aggregator only increases from 0).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Restructure the docs ahead of sharing the project for others to self-host. The
README is now a focused get-it-running path (overview, architecture, prereqs,
quick start, operational notes, security, backup/restore). Moved the fuller
feature tour, the maintainer pre-built-image release flow, and the Gluetun VPN
guide into a new docs/DEPLOYMENT.md, cross-linked both ways. The pre-built-image
section is now labelled a maintainer process with a LYRA_REGISTRY override so a
self-hoster can point at their own registry (or just build from source).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Previously the /lastfm rows only showed a status chip; following an artist or
wanting an album meant opening the modal first. Add an inline Follow button to
each Artists-tab row and an inline Want button to each Albums-tab row, reusing
the same endpoints as the modals (resolve MB id/release-group as needed, then
POST /api/artists or /api/discover/want). Optimistic per-row state flips the
button to the Following/Wanted chip without a reload; the button is disabled
while the request is in flight.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A track beyond the MusicBrainz tracklist (a Qobuz bonus track) kept streamrip's
raw name (e.g. "13. John Mayer - St. Patrick's Day (Album Version).flac") while
the main tracklist got clean "## Title" names. plan_track_names now canonicalizes
such extras: strip the leading track-number prefix and a leading "<artist> - ",
yielding "13 St. Patrick's Day (Album Version).flac". Requires an explicit
delimiter after the number so titles like "99 Luftballons" are left alone; with
no tracklist resolved at all the on-disk stem is still kept.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Soulseek's slskd search polls up to ~2 min; searching adapters sequentially
added that on top of Qobuz's instant search. Run the per-adapter searches
concurrently via a ThreadPoolExecutor so total match time is the slowest
single source, not their sum. Only Qobuz drives streamrip's shared asyncio
loop and there's one Qobuz adapter, so no two threads use that loop at once.
Results collected in adapter order (map preserves order) for deterministic
candidate ordering; per-adapter failures still swallowed (down source = no
candidates).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
streamrip reports progress only per concurrent track, so Job.downloadProgress
came solely from the file-count poller — 0% until every track landed, then a
jump to 100%. Patch streamrip.media.track.get_progress_callback to aggregate
per-track byte deltas over an estimated album total (avg started-track size ×
len(album.tracks)) into one smooth, generally-climbing 0..1 fraction routed
through the pipeline's on_progress. Capped at 0.99; the pipeline sets 1.0 on
completion. No signature changes — track count read from the resolved album.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Qobuz progress smoothing, parallel source searches, bonus-track filename
canonicalization. Lyra is live on the VM; these are worker-only polish.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds scripts/build-push.sh (build + push lyra-web/lyra-worker :latest and
:<sha> to git.jger.nl) + a README 'Deploying to a server (pre-built images)'
section. The server pulls these via the vm-download ops repo (docker/lyra).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A 4xx/5xx (rate limit / auth) now raises instead of parsing an error body;
callers already handle it (discovery per-source try/except; album fetch
returns []). Closes the last small deferred cleanup.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ListenBrainz surfaces producers/session players/band members (Max Martin,
Dominic Howard) that co-occur in listens but have no releases of their own —
they showed as bare rows with an empty page. _derive_albums now drops a
surfaced artist (deletes their pending suggestion) when their MB discography
has no core release-group (Album/Single/EP, no secondary type), catching both
empty discographies and soundtrack/production-only credits, while keeping real
singles/EP-only artists. On browse failure the suggestion is kept.
Test fake browser gains default_core so scoring tests keep surfacing candidates.
worker 249 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
POST /api/discover/reset sets discover.resetRequested; the worker, at the
start of the next sweep, clears all PENDING suggestions + contributions +
seed throttles and nulls WatchedArtist.lastDiscoveredAt (keeping
followed/wanted/dismissed decisions), then regenerates everything with current
logic. A 'Reset' button on /discover (2-step confirm) triggers it.
worker 247, web 190 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Client filters the row's album thumbnails to full albums only (hides
singles/EPs left over from pre-albums-only sweeps).
- Worker: re-deriving an artist deletes its pending album suggestions that are
no longer picks (stale singles / now-owned), keeping wanted/dismissed rows,
so the suggestion set self-cleans instead of accumulating.
worker 246, web 189 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each suggested album is now a small horizontal item (30px cover + title/badge
on one line, like Recently Pressed but smaller) stacked vertically, so both
albums fit under each other without enlarging the row.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
/discover is now one list of artist rows (no Artists|Albums tabs). Each row
shows the artist (name links to the full page — ArtistModal dropped from
Discover), the 'Similar to' reason, and its newest/most-played album
thumbnails with badges. Album-click opens the AlbumModal (cover + tracklist +
Want). Find-similar results link to the full page too.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
/api/discover now returns rows[] grouped by artist, each carrying its album
suggestions (with albumReason) + the seed names, ordered by score. Album
suggestions whose artist has no pending artist suggestion still yield a row.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
_derive_albums now also picks the artist's most-played album (Last.fm,
matched into the browsed discography by MBID then normalized title),
alongside the newest. Same release -> one row tagged newest,popular.
albumReason persisted; run_discovery threads an optional popularity provider.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
LastfmAlbumPopularity.top_albums(conn, name) ranks an artist's albums by
Last.fm playcount (read-through ApiCache, 12h; errors -> []). registry
build_album_popularity returns it only when lastfm.api_key is set.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Design for collapsing the Artists|Albums tabs into one artist-centric row,
surfacing each similar artist's newest AND most-popular (Last.fm) album inline,
album-click → AlbumModal tracklist, artist name → full page (drop the Discover
ArtistModal). Approved design; implementation plan next.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Discovery seeded only from followed artists. Now it also seeds from artists
you own albums by (LibraryItem.artistMbid, captured since the MBID own-state
work) but don't follow — so your library shapes suggestions, not just your
follows.
- New DiscoverySeed throttle table (migration add_discovery_seed) mirroring
WatchedArtist.lastDiscoveredAt, so each library-derived seed is swept once
per interval (WatchedArtist seeds keep their own throttle).
- _seed_artists merges both seed sources least-recently-swept first, capped at
chunk_size; run_discovery stamps the right throttle per seed and evicts
contributions from seeds that are neither followed nor owned.
- Artists you already have (followed OR owned) are excluded from the suggestion
candidates — discovery surfaces only genuinely new artists.
- discover.seedFromLibrary config (default true) + a Discovery settings toggle.
NOTE: existing LibraryItem rows have null artistMbid (populated on new
import/scan), so a re-scan is needed before old albums seed discovery.
worker 236, web 187 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each suggestion now carries the followed artists (seeds) that surfaced it,
from DiscoverySeedContribution joined to the current watched roster (an
unfollowed seed drops out), highest-scoring seed first. /discover renders
"Similar to Arctic Monkeys, Coldplay +2" under each artist and album row —
both kinds key off the candidate's artistMbid, so an album inherits its
artist's seeds. web 187 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Own-state ("in library") relied on an exact artist-NAME match, so an owned
album whose name differed from the MB canonical (punctuation/locale/feat.)
showed as not-owned even when followed. Capture the MusicBrainz artist MBID
on LibraryItem and prefer it for matching.
- New LibraryItem.artistMbid (migration add_library_artist_mbid). Populated at
import (pipeline) + scan, both of which already hold the resolved target;
the pipeline also now stores rgMbid (it previously didn't). COALESCE on
conflict backfills MBIDs on re-scan without clobbering.
- /api/artists "in library, not followed" matches by artistMbid first, name
case-insensitively as fallback (rows predating the column). Library route
prefers the item's own artistMbid over the release join.
Last.fm/discover own-state still use name-match (they work) — a smaller
follow-up. worker 233, web 186 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The verify step checked track COUNT but not playtime, so a truncated file or
a short preview substituted for a real track could import as "complete". Add
a total-duration check: sum the staged audio's playtime (mutagen) and fail
the job when it falls below 85% of MusicBrainz's expected total
(target.total_duration_s, already resolved). Generous tolerance so
edition/encoding differences don't false-positive; bonus tracks (over-long)
always pass.
The duration reader is injected (measure_duration param, default the real
mutagen-based reader) so the fail/pass paths are unit-tested; it returns None
for the offline fakes (no real audio), skipping the check exactly like the
staged-count guard. worker 233 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
_record_contributions now normalizes each source's top-N to [0,1] by its own
top score before summing across sources. ListenBrainz co-occurrence counts
(thousands) no longer drown Last.fm's 0–1 match scores, so a small-scale
source contributes real ranking signal instead of only adding candidates.
Discovery score assertions updated to the normalized scale (single-candidate
fakes normalize to 1.0); added a cross-source test showing a small-scale
source's candidate ranks equal to a large-scale one. min_score is now on the
[0, n_sources] scale (default 0.0 → unaffected). worker 231 tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Jonathan