fix: circuit breaker recovery, recordSuccess race, vibe queue self-heal

Fix 1 — Circuit breaker deadlock (audioAnalysisCleanup.ts):
shouldAttemptReset() was measuring time since last failure, which resets
every cleanup cycle. Added circuitOpenedAt fixed at the moment the breaker
first opens. shouldAttemptReset() now measures from that fixed point so the
5-minute window actually expires regardless of ongoing failure activity.

Fix 2 — recordSuccess() race condition (unifiedEnrichment.ts):
recordSuccess() compared audioCompletedBefore/After bracketing only
cleanupStaleProcessing() — a millisecond window that never captured Python
completions (~14s batch cadence → effectively 0% hit rate). Replaced with
audioLastCycleCompletedCount tracked across cycles. executeAudioPhase() now
fires recordSuccess() whenever completed count grew since the previous cycle
(~1 min window), reliably detecting analyzer progress.

Fix 3 — CLAP vibe embeddings queue self-healing (unifiedEnrichment.ts):
queueVibeEmbeddings() filtered vibeAnalysisStatus = 'pending', silently
skipping 7286 tracks left as 'completed' after reduce_embedding_dimension
migration dropped their embeddings. Changed filter to <> 'processing' so
te.track_id IS NULL (actual embedding existence) is the source of truth.
Queue now self-heals if embeddings are ever wiped again.

Author: Your Name

backend/src/services/audioAnalysisCleanup.ts

@@ -13,11 +13,12 @@ class AudioAnalysisCleanupService {
     private state: CircuitState = "closed";
     private failureCount = 0;
     private lastFailureTime: Date | null = null;
+    private circuitOpenedAt: Date | null = null;
 
     private shouldAttemptReset(): boolean {
-        if (!this.lastFailureTime) return false;
-        const timeSinceFailure = Date.now() - this.lastFailureTime.getTime();
-        return timeSinceFailure >= CIRCUIT_BREAKER_WINDOW_MS;
+        if (!this.circuitOpenedAt) return false;
+        const timeSinceOpen = Date.now() - this.circuitOpenedAt.getTime();
+        return timeSinceOpen >= CIRCUIT_BREAKER_WINDOW_MS;
     }
 
     private onSuccess(): void {
@@ -28,12 +29,14 @@ class AudioAnalysisCleanupService {
             this.state = "closed";
             this.failureCount = 0;
             this.lastFailureTime = null;
+            this.circuitOpenedAt = null;
         } else if (this.state === "closed" && this.failureCount > 0) {
             logger.debug(
                 "[AudioAnalysisCleanup] Resetting failure counter on success"
             );
             this.failureCount = 0;
             this.lastFailureTime = null;
+            this.circuitOpenedAt = null;
         }
     }
 
@@ -48,11 +51,14 @@ class AudioAnalysisCleanupService {
 
         if (this.state === "half-open") {
             this.state = "open";
+            // Don't reset circuitOpenedAt — keep original open time so the
+            // next shouldAttemptReset() fires immediately and retries HALF-OPEN
             logger.warn(
                 `[AudioAnalysisCleanup] Circuit breaker REOPENED - recovery attempt failed (${this.failureCount} total failures)`
             );
         } else if (this.failureCount >= CIRCUIT_BREAKER_THRESHOLD) {
             this.state = "open";
+            this.circuitOpenedAt = new Date(); // fixed point — not updated on subsequent failures
             logger.warn(
                 `[AudioAnalysisCleanup] Circuit breaker OPEN - ${this.failureCount} failures in window. ` +
                     `Pausing audio analysis queuing until analyzer shows signs of life.`
@@ -80,6 +86,7 @@ class AudioAnalysisCleanupService {
         this.state = "closed";
         this.failureCount = 0;
         this.lastFailureTime = null;
+        this.circuitOpenedAt = null;
         logger.debug("[AudioAnalysisCleanup] Circuit breaker reset");
     }
 

backend/src/workers/unifiedEnrichment.ts

@@ -41,6 +41,7 @@ let isStopping = false;
 let immediateEnrichmentRequested = false;
 let consecutiveSystemFailures = 0; // Track consecutive system-level failures
 let lastRunTime = 0;
+let audioLastCycleCompletedCount: number | null = null;
 const MIN_INTERVAL_MS = 10000; // Minimum 10s between cycles
 
 // Batch failure tracking
@@ -987,7 +988,7 @@ async function queueVibeEmbeddings(): Promise<number> {
           LEFT JOIN track_embeddings te ON t.id = te.track_id
           WHERE te.track_id IS NULL
             AND t."filePath" IS NOT NULL
-            AND (t."vibeAnalysisStatus" IS NULL OR t."vibeAnalysisStatus" = 'pending')
+            AND (t."vibeAnalysisStatus" IS NULL OR t."vibeAnalysisStatus" <> 'processing')
           LIMIT 1000
       `;
 
@@ -1072,10 +1073,17 @@ async function executeMoodTagsPhase(): Promise<number> {
 }
 
 async function executeAudioPhase(): Promise<number> {
-    const audioCompletedBefore = await prisma.track.count({
+    // Compare completed count to previous cycle (~1 min ago) — a much wider
+    // window than the milliseconds between two counts around cleanupStaleProcessing().
+    const currentCompleted = await prisma.track.count({
         where: { analysisStatus: "completed" },
     });
 
+    if (audioLastCycleCompletedCount !== null && currentCompleted > audioLastCycleCompletedCount) {
+        audioAnalysisCleanupService.recordSuccess();
+    }
+    audioLastCycleCompletedCount = currentCompleted;
+
     const cleanupResult =
         await audioAnalysisCleanupService.cleanupStaleProcessing();
     if (cleanupResult.reset > 0 || cleanupResult.permanentlyFailed > 0) {
@@ -1084,13 +1092,6 @@ async function executeAudioPhase(): Promise<number> {
         );
     }
 
-    const audioCompletedAfter = await prisma.track.count({
-        where: { analysisStatus: "completed" },
-    });
-    if (audioCompletedAfter > audioCompletedBefore) {
-        audioAnalysisCleanupService.recordSuccess();
-    }
-
     if (audioAnalysisCleanupService.isCircuitOpen()) {
         logger.warn(
             "[Enrichment] Audio analysis circuit breaker OPEN - skipping queue",