fix: dedupe and cap concurrent lyrics plugin fetches (#5792)

* fix: dedupe and cap concurrent lyrics plugin fetches

Clients like Finamp prefetch lyrics for several queue tracks at once. The
resulting burst of concurrent plugin calls can rate-limit the primary
lyrics provider into a timeout, making the plugin fall back to a lower
quality source and cache the bad result.

SimpleCache.GetWithLoader now deduplicates concurrent loads of the same
key via singleflight, with every waiter receiving the winner's result or
error. The Jellyfin lyrics loader is detached from the request context so
one cancelled request cannot fail the load for all waiters, and the
lyrics adapter caps in-flight plugin calls at 2 per plugin, queueing the
rest. As a side effect, the cached HTTP client used by the Last.fm,
Deezer and ListenBrainz agents also collapses identical concurrent
requests into a single upstream call.

* fix: harden lyrics concurrency fixes per review

Replace the stringified singleflight keys with a per-cache flight map
keyed by the cache key type itself, eliminating potential key collisions
for non-string keys, the nil-interface assertion panic, and the
stringification overhead. Release the lyrics semaphore slot via defer so
a panicking plugin call cannot leak it, and bound the detached lyrics
load with a one-minute timeout so a hung plugin cannot pin its
singleflight and semaphore slot indefinitely.

server/jellyfin/lyrics.go

@@ -10,11 +10,18 @@ import (
 	"github.com/navidrome/navidrome/server/jellyfin/dto"
 )
 
+const lyricsLoadTimeout = time.Minute
+
 // cachedLyrics resolves lyrics through the full source pipeline (embedded, sidecar, plugins),
 // caching results — including empty: clients poll per played track, so misses are the hot path.
 func (api *Router) cachedLyrics(ctx context.Context, mf *model.MediaFile) model.LyricList {
+	// The load is shared across requests (singleflight) and cached, so don't let one
+	// cancelled request abort it for everybody — detach it from the request's lifetime,
+	// keeping a bound so a hung plugin can't pin the fetch (and its plugin slot) forever.
+	loadCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), lyricsLoadTimeout)
+	defer cancel()
 	list, err := api.lyricsCache.GetWithLoader(mf.ID, func(string) (model.LyricList, time.Duration, error) {
-		l, err := api.lyrics.GetLyrics(ctx, mf)
+		l, err := api.lyrics.GetLyrics(loadCtx, mf)
 		return l, 0, err // 0 → cache DefaultTTL
 	})
 	if err != nil {

server/jellyfin/lyrics_test.go

@@ -17,13 +17,18 @@ import (
 
 // fakeLyricsService returns canned lyrics per media-file ID and counts calls.
 type fakeLyricsService struct {
-	lyrics map[string]model.LyricList
-	err    error
-	calls  int
+	lyrics      map[string]model.LyricList
+	err         error
+	calls       int
+	hadDeadline bool
 }
 
-func (f *fakeLyricsService) GetLyrics(_ context.Context, mf *model.MediaFile) (model.LyricList, error) {
+func (f *fakeLyricsService) GetLyrics(ctx context.Context, mf *model.MediaFile) (model.LyricList, error) {
 	f.calls++
+	_, f.hadDeadline = ctx.Deadline()
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
 	if f.err != nil {
 		return nil, f.err
 	}
@@ -129,4 +134,22 @@ var _ = Describe("getLyrics", func() {
 		Expect(doRequest("s2").Code).To(Equal(http.StatusNotFound))
 		Expect(fake.calls).To(Equal(1))
 	})
+
+	It("completes and caches the fetch even when the request context is cancelled", func() {
+		fake.lyrics["s1"] = model.LyricList{
+			{Kind: "main", Synced: true, Line: []model.Line{{Start: p(1000), Value: "hello"}}},
+		}
+		ctx, cancel := context.WithCancel(context.Background())
+		cancel()
+
+		list := api.cachedLyrics(ctx, &model.MediaFile{ID: "s1"})
+		Expect(list).ToNot(BeEmpty())
+		Expect(doRequest("s1").Code).To(Equal(http.StatusOK))
+		Expect(fake.calls).To(Equal(1))
+	})
+
+	It("bounds the detached fetch with a timeout", func() {
+		Expect(doRequest("s2").Code).To(Equal(http.StatusNotFound))
+		Expect(fake.hadDeadline).To(BeTrue())
+	})
 })