The cross-user bucket query combined SELECT DISTINCT with ORDER BY md5(...),
which Postgres rejects at plan time (SQLSTATE 42P10). buildDiscoverCandidates
returned that error on first bucket failure, so the whole Discover playlist
was skipped every nightly run — even though the random bucket pool was
healthy. Switched to GROUP BY so the md5 ordering expression no longer needs
to appear in the select list, and hardened the function so future single-
bucket failures degrade gracefully via slot redistribution instead of taking
out the whole playlist.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two unrelated wins as a single batch.
Flutter — lazy source building in setQueueFromTracks:
Today: Future.wait builds all N AudioSource objects (drift queries +
LockCaching ctor) before the player can call setAudioSources →
play(). Measured at 83ms for a 25-track playlist, on top of the
~285ms initial-source preload.
New flow: build only the initial source, hand it to setAudioSources
([initial], initialIndex: 0) so play() can start, then background-
fill the rest. Forward direction (skipNext targets) added via
addAudioSource. Backward direction (skipPrev) inserted at index 0..
initialIndex-1 with _suppressIndexUpdates true so the unavoidable
currentIndex shifts don't push the wrong MediaItem onto the stream.
Saves the up-front source-build wait — tap-to-audio for long queues
should drop by ~80-100ms even on cache hits.
Server — Cache-Control on the three byte-serving endpoints:
- /api/albums/{id}/cover: max-age=86400, must-revalidate. Covers
change rarely (re-scan, MBID enrichment); a day of cache is safe
and skips conditional GETs for the bulk of a session.
- /api/playlists/{id}/cover: max-age=300, must-revalidate. Collages
recompute when contents change; short enough for edits to feel
fresh, long enough to skip repeat fetches during a session.
- /api/tracks/{id}/stream: max-age=31536000, immutable. Track bytes
are immutable for a given id (scanner re-indexes by file_path; new
files get new ids). LockCachingAudioSource on the Flutter side
already disk-caches, but proper headers let it skip even the
conditional 304 on repeat plays.
handleGetHome itself is well-architected (5 sections in parallel via
goroutines, latency-bound by the slowest single query). The cold-
start lag is two of those queries doing wider scans than necessary.
ListLastPlayedArtistsForUser was iterating FROM artists a with a
LATERAL play_events join per row — O(total_artists in library) plan
even for users who've only played a handful. Inverted: aggregate the
user's plays by artist_id first via the play_events → tracks join
(uses play_events_user_track_idx + tracks pkey), then attach the
artist row and lateral cover/count subqueries only for the artists
that actually appear. Cost now bounded by play history, not library
size.
ListMostPlayedTracksForUser was joining tracks/albums/artists for
every play_event row before grouping — O(total plays) work for
joins. Pre-aggregated play_events into a CTE keyed by track_id +
count(*), then joined to tracks/albums/artists only for the
distinct-tracks survivors. Order-by uses the pre-computed count.
No handler or generated-Go signature changes — both queries return
the same rowset shape, just much faster on libraries where total
artists/plays >> distinct-played-artists/distinct-played-tracks.
Two new sqlc queries replace three sequential per-album round trips
that were dominating detail-screen latency.
GetAlbumWithArtist: handleGetAlbum was doing GetAlbumByID then
GetArtistByID — separate round trips for one logical lookup. The new
query joins albums + artists with sqlc.embed and returns both in one
SELECT. Detail-page DB cost: 3 trips → 2.
ListAlbumsByArtistWithTrackCount: handleGetArtist was loading the
artist's album list, then issuing one CountTracksByAlbum per album to
populate track_count. On a 30-album artist that's 32 sequential
queries — each ~5ms over a local DB, ~30ms over a remote one. The
new query embeds the album row + a correlated count(*) subquery, so
every album's track count comes back in one SELECT regardless of
album count. Detail-page DB cost: 1 + N → 1 + 1.
Together these account for the bulk of cold-cache navigation latency
on the Flutter client. Combined with the existing SWR + nav
hydration on the client side, detail screens should render their
header instantly and the body within one round trip instead of
N+constant.
Came up debugging the in-app update flow — wasn't obvious which image
the container was actually running without exec'ing in. New flow:
### Server
- internal/server/version.go: new `var ServerVersion = "dev"`,
overridden via -ldflags at build time.
- /healthz response gains a "version" key alongside the existing
"status" + "min_client_version". Backward-compat: existing clients
ignore unknown JSON fields. Endpoint stays unauthenticated.
### Build
- Dockerfile: new `ARG MINSTREL_VERSION=dev`, threaded into the
go build -ldflags so the binary's ServerVersion is stamped at
link time. Default "dev" preserves local `docker build` ergonomics.
- .forgejo/workflows/release.yml: tags step also emits a `version`
output (the git tag for tag pushes, "main" for branch pushes);
build step passes it as `--build-arg MINSTREL_VERSION=...`.
### Web
- web/src/lib/components/ServerVersion.svelte: small understated
text ("Server v2026.05.10.2") that fetches /healthz on mount.
Renders nothing on parse failure or pre-version images.
- Mounted at the bottom of Settings.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes the bandwidth-abuse vector on the in-app update flow. Both
endpoints now sit inside the authed.Group; APK additionally gets a
60s/user rate limit to suppress accidental hammering or scripted
abuse.
### Server
- internal/api/client_assets.go:
- clientAPKAllowDownload(): in-memory map[userID]time.Time under
a mutex. Returns 0 (allow) or wait duration (block).
- handleClientAPK reads user from context, checks the limit,
returns 429 + Retry-After header if blocked.
- testResetClientAPKRateLimit() lets tests start clean.
- internal/api/api.go: routes moved from the root /api group into
the authed.Group block (alongside /quarantine, /requests, etc.).
- Tests: added TestClientAPK_401WhenUnauthenticated and
TestClientAPK_RateLimit_429OnRapidSecondCall (also verifies
different user gets a fresh slot). Existing tests updated to use
authedRequest() helper.
### Web
- MobileAppDownload.svelte: switched from bare fetch (no credentials)
to api.get<>() which carries the session cookie. 404 / 401 /
network errors all silently hide the download row.
- Removed the login-page mount entirely — pre-auth surfaces should
never show this. Settings → Mobile app section keeps it for
logged-in users.
Flutter unaffected: dio's Bearer interceptor already attaches the
token, and the polling only fires once the post-login shell mounts
the banner widget.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- flutter analyze: Riverpod 3 dropped StateProvider; replaced
_dismissedVersionsProvider with a small Notifier<Set<String>>
+ NotifierProvider, mutating via an `add(version)` method.
Public API (shouldShowUpdateBannerProvider, dismissUpdateProvider)
unchanged.
- golangci errcheck: defer f.Close() now wrapped in func() { _ = f.Close() }();
os.Setenv calls in test helper switched to t.Setenv (cleaner — auto-restores
on test cleanup, no manual Unsetenv needed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 1 of the in-app update flow — server side. Endpoints serve the
bundled Android APK + sidecar version file from /app/client/.
Returns 404 gracefully when files aren't present, so dev environments
and pre-CI-wiring images degrade cleanly to "no update available."
- internal/api/client_assets.go: handleClientVersion + handleClientAPK.
Both unauthenticated (matches /healthz) so install flow doesn't
depend on a live session. APK served with proper
application/vnd.android.package-archive Content-Type +
http.ServeContent so Range requests work for resumable downloads
on flaky networks.
- Path resolves to /app/client/ by default; MINSTREL_CLIENT_APK_DIR
env var overrides for dev.
- Dockerfile creates /app/client/ + commented COPY hooks for the CI
sequencing phase.
- Tests cover all four states: missing apk, apk-but-no-version,
both present (200 with correct shape), apk stream (200 with
correct Content-Type + body bytes).
Phases 2 (Flutter client provider + banner + install intent + Android
manifest changes) and 3 (CI sequencing to bake the APK into the image)
land in follow-up commits.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two fixes in one commit because they're entangled — the systemVariant
work would have been theater otherwise.
## The wire-format bug
/api/library/sync was emitting PascalCase JSON for artist / album /
track / playlist upserts (raw json.Marshal of sqlc-generated structs
with no JSON tags — sqlc.yaml: emit_json_tags=false). Flutter's
sync_controller _*FromJson reads snake_case keys, so all metadata
sync rows landed in drift with empty strings / zero ints.
The like_track / like_album / like_artist / playlist_track entities
work because they're hand-built `map[string]string` payloads with
snake_case keys — they sidestepped the bug. The 4 raw-marshal
entities did not.
Existing sync test caught zero of this — it asserts on len(upserts)
not field shape.
Fix: server-side view structs in library_sync_views.go with proper
JSON tags + pgtype-flattening (UUID → 8-4-4-4-12 hex string,
Date → "2006-01-02"). Mirrors the playlistRowView pattern from
/api/playlists. New library_sync_views_test.go pins the wire keys
so future field-name drift breaks loud.
## systemVariant column (closes#357 plan C v1 limitation)
playlistSyncView now carries `system_variant` server → wire.
Flutter drift schema bumped from 1 → 2 with onUpgrade adding the
`systemVariant TEXT NULL` column to cached_playlists. Cursor reset
to 0 in the migration so existing rows refresh with the new field
on the next sync.
playlistsListProvider now filters locally by systemVariant:
- kind='user' → systemVariant IS NULL (the add-to-playlist sheet's intent)
- kind='system' → systemVariant IS NOT NULL
- kind='all' → no filter
Closes the documented v1 limitation where the add-to-playlist sheet
showed system playlists alongside user-created ones.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
golangci-lint surfaced both on 9c7dec6:
- errcheck on bare 'defer tx.Rollback(ctx)' in 2 test files
- gofmt -s wanted tighter map-key alignment in library_sync.go +
library_sync_test.go (auto-fixed by 'gofmt -s -w')
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Five sites wired:
- Create / Update / Delete playlist: pool-bound LogChange after success
- AppendTracks: tx-bound LogChange per inserted playlist_track row
- RemoveTrack: tx-bound LogChange after the delete + lookup the
track_id pre-delete so the composite key is still resolvable
The two tx-bound paths (AppendTracks, RemoveTrack) treat LogChange as
required — failure rolls back the playlist mutation too. The pool-bound
paths Warn on failure to match the scanner / likes pattern (mutation
already committed; missed log row recovers on next mutation).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
All 6 like/unlike sites (track/album/artist × like/unlike) now emit a
library_changes row via the shared logLikeChange helper. Best-effort:
LogChange failures Warn but don't fail the HTTP response — a missed
log row is recovered at the next mutation on the same entity.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wires sync.LogChange into the library mutation sites so /api/library/sync
reflects upserts and deletes.
Architectural pivot: LogChange's signature is now (ctx, dbq.DBTX, ...) so
it works with both *pgxpool.Pool and pgx.Tx. The scanner doesn't run
mutations in explicit transactions, so it pool-binds; delete.go matches.
Tx-bound callers (likes/playlists in subsequent commits) keep atomicity.
Also: sync.FormatUUID centralizes the pgtype.UUID → canonical string
conversion that both the scanner and the sync handler need; library_sync.go
now uses it instead of a local copy.
Best-effort logging on scanner failures (Warn, don't fail the scan): a
LogChange error after a successful upsert is rare and self-healing — the
next scan that touches the entity re-emits the change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Returns batched upserts + deletes since the supplied cursor. Empty cursor
returns full snapshot; subsequent calls pull deltas. Per-user entities
(likes, playlists) are scoped to the authed user. Composite-key entities
(likes, playlist_tracks) use stable string ids encoded by sync.EncodeLikeID
/ sync.EncodePlaylistTrackID.
Behavior:
204 No Content - no changes since cursor
200 OK - JSON syncResponse {cursor, upserts, deletes}
410 Gone - cursor older than oldest log row; client must reset
401 / 500 - standard envelope errors
Adds sqlc queries GetAlbumsByIDs, GetTracksByIDs, GetPlaylistsByIDs to
mirror the existing GetArtistsByIDs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New package internal/sync. LogChange writes a library_changes row inside
the supplied tx. Encode helpers produce stable composite ids for like_*
and playlist_track entries. Subsequent commits wire LogChange into the
scanner / likes / playlists services.
Also: dbtest.dataTables now includes library_changes so test isolation
holds across runs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Append-only change log for library entities. Every mutation on
artists/albums/tracks/likes/playlists/playlist_tracks will write a row
in the same transaction as the mutation itself (wired in subsequent
commits). Powers the Flutter delta-sync endpoint (#357).
- 0025_library_changes migration (up + down)
- internal/db/queries/library_changes.sql (Insert, GetSince, MaxCursor, MinCursor)
- regenerated dbq from sqlc
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Updates the happy-path test stub to handle the three list endpoints
Lidarr exposes (qualityprofile / metadataprofile / rootfolder) and
asserts profiles + folders make it through to the response.
- Adds a partial-failure case where one list endpoint 5xxs; verifies
ok=true, the failing list is omitted, and list_errors carries its
bucket code.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extends POST /api/admin/lidarr/test response to include quality_profiles,
metadata_profiles, root_folders, and an optional list_errors map when the
Lidarr connection succeeds. The three list fetches run in parallel after
the ping; any per-list failure goes into list_errors so the response can
still report ok=true with whatever data did come back. Failed-ping
response shape is unchanged.
This lets /admin/integrations populate its dropdowns on a single
round-trip during first-time setup, fixing the chicken-and-egg where
the dropdowns previously gated on cfg.Enabled (which can't be true
until the first save, which itself needs non-zero defaults).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The bootstrap-admin-from-env-vars flow was a holdover from before
self-registration could create the first admin. Now that handleRegister
+ CreateUserFirstAdminRace promotes the first user to register on an
empty users table (verified shipped in v2026.05.08.2 / #376), the
bootstrap path is just a second source of truth that confuses operators
(and leaves an "admin" account in the DB that nobody asked for).
Removes:
- internal/auth/bootstrap.go + its test
- The auth.Bootstrap call from cmd/minstrel/main.go
- AuthConfig + AdminBootstrapConfig structs from config
- MINSTREL_AUTH_ADMIN_USERNAME / _PASSWORD env reads + their test
- The auth: block from config.example.yaml
- Bootstrap-related comments in docker-compose.yml + README.md
The README quickstart now points operators at /register on first start
instead of "watch the logs for a one-time password."
Existing instances keep their bootstrap-admin row in the DB; operators
who want it gone can register a new admin via /register, promote them
in /admin/users, then delete the old bootstrap user (last-admin guard
will require the new admin to be promoted first). No migration needed.
Recovery story for forgotten admin passwords now hinges on Fable #321
(admin password reset CLI) — currently the only path back in if no
other admin exists.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add audit.WriteOrLog: a one-line wrapper around Write that logs at
Warn and swallows the error, matching the package contract that
audit failures must not break user-facing operations.
Migrate the 13 call sites across 7 files in internal/api/ from the
3-line "if err != nil { logger.Warn(...) }" shape to a single call.
audit.Write stays exported for tests + any future caller that
needs strict semantics.
Adds three tests: success (no log), failure-via-closed-pool (Warn
record with action+err keys), and nil-logger (no panic). Tests
skip when MINSTREL_TEST_DATABASE_URL is unset, matching the
existing harness convention.
The PR1-T2 admin handler migration replaced bespoke 500-class messages
("lookup failed", "refetch failed", "trigger failed", "bump failed",
"remove failed", "update failed", "schedule row missing", "re-read
failed", "read failed") with apierror.Internal(err), which forces the
wire Message to "internal server error". That violates the PR1
contract that errEnvelope{Code, Message} must remain byte-identical
for existing clients.
Add apierror.InternalMsg(message, cause) — a 500-class constructor
that preserves the call site's user-facing message while keeping the
cause attached for logging and errors.Is. Re-migrate the 12 admin
sites whose pre-1cc7eb6 source had a non-empty bespoke Message so
they restore the original wire string. Sites whose original Message
was empty stay on Internal(err) (humanization to "internal server
error" is a tolerable improvement, not a regression).
admin_smtp.go's send_failed site (line 129) was already preserved
via a raw &apierror.Error literal in 1cc7eb6 and needs no fix.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Rewrite writeErr(w, err) to wrap *apierror.Error via apierror.From,
preserving the existing {"error": {"code", "message"}} wire envelope.
Add writeErrWithLog helper for 500-class errors that need an operator
log line. Migrate all 13 admin_*.go handler files (~76 call sites) to
the new signature; T3 will sweep the remaining api package.
The old 4-arg writeErr is removed, so non-admin call sites in
internal/api will not compile until T3 lands. This is by design — T2
and T3 are paired.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wires the auto_approve_requests user flag into the request creation
flow. When the flag is set (admin enabled it via /admin/users from
U2-T3), POST /api/requests transitions the just-created pending row
through Service.Approve inline, which dispatches to Lidarr the same
way a manual admin approve does.
Failures inside Approve (ErrLidarrDisabled, ErrDefaultsIncomplete,
network/Lidarr errors) leave the row pending — same fallback as
manual approve hitting the same path. The user gets a 201 either
way; admin can still resolve the row in /admin/requests if it
stayed pending. The handler logs the auto-approve failure with
user_id + request_id for observability.
The "actor" id passed to Approve is the user's own ID — they're
acting under the privilege the admin granted them via the toggle.
The trail of "admin set the flag" already lives in audit_log
(ActionAutoApproveToggle from U2-T2). Adding a per-auto-approval
audit entry is a future enhancement; for v1 the toggle audit plus
the request row's status transition is enough.
Test verifies the Lidarr-disabled fallback contract: a user with
auto_approve=true and Lidarr unavailable still gets 201 Created
with status=pending (no crash, no 500). The "auto-approve actually
succeeds" test path requires a Lidarr stub; deferred until a
broader Lidarr test fixture lands.
Two real algorithm bugs in F-T1's For-You composition + Discover
allocator. Both surfaced as failing unit tests under go test -race.
1. redistributeSlots was re-redistributing a bucket's full deficit
on every pass instead of just the residual. The loop computed
`deficit = b.want - final[i]` each iteration, but final[i] for
a deficit bucket never increases (its supply is exhausted), so
pass N saw the same deficit as pass N-1 and kept shoveling it
to peers. For [want:40 avail:100, want:30 avail:0, want:30 avail:100],
four passes pushed cross-user's deficit into dormant+random four
times each, hitting the 100-slot clamp at the end and producing
[50, 0, 50] instead of the spec'd [55, 0, 45].
Fix: track per-source `redistributed[i]` and subtract it from the
deficit each pass. Multi-pass behavior still works for the case
where a peer's supply runs out mid-distribution.
2. tieBreakHash used FNV-1a 64-bit with trackID + dateStr appended.
For dateStrs differing only in the last character ("2026-05-07"
vs "2026-05-08"), the FNV state diverged only in low bits at the
final byte; multiplication by FNV_prime propagates upward but the
relative ordering of 60 small candidate UUIDs (which differ only
in their last byte) ended up identical across the two dates. The
For-You head/tail test asserted that the tail's first 5 should
change across days; it didn't.
Fix: switch to SHA-256 truncated to 8 bytes. SHA-256 has full
avalanche, so any single-bit input change roughly half-flips the
output bits and meaningfully reorders.
The hash isn't security-load-bearing; we just need strong avalanche
for tiny dateStr deltas. Determinism (same inputs → same output) is
preserved.
bvandeusen