fix: users can retriger transcription

apps/jobs/src/workers/maintenance.ts

@@ -1,12 +1,13 @@
 import './../bootstrap';
 
-import { Worker } from 'bullmq';
+import { Queue, Worker } from 'bullmq';
 
-import { pruneDeadLetterJobsOlderThan } from '@cio/db/queries';
+import { pruneDeadLetterJobsOlderThan, reapStuckMediaJobs } from '@cio/db/queries';
 import {
   JOB_NAMES,
   QUEUE_NAMES,
   ZDeadLetterCleanupPayload,
+  ZMediaJobReapPayload,
   ZRetentionCompactPayload,
   createRedisConnection
 } from '@cio/jobs';
@@ -36,11 +37,56 @@ const worker = new Worker(
       return { compacted: 0 };
     }
 
+    if (job.name === JOB_NAMES.maintenance.mediaJobReap) {
+      const data = ZMediaJobReapPayload.parse(job.data ?? {});
+      const cutoffIso = new Date(Date.now() - data.staleAfterMinutes * 60 * 1_000).toISOString();
+      const reaped = await reapStuckMediaJobs(cutoffIso);
+
+      if (reaped.length > 0) {
+        log.warn('media-job-reap-done', {
+          reaped: reaped.length,
+          cutoffIso,
+          ids: reaped.map((row) => row.id)
+        });
+      }
+
+      return { reaped: reaped.length };
+    }
+
     throw new Error(`Unknown maintenance job: ${job.name}`);
   },
   { connection, concurrency: 1 }
 );
 
+/**
+ * BullMQ repeatable scheduler for the media-job reaper. `upsertJobScheduler`
+ * is idempotent on the scheduler id, so it's safe to call on every worker
+ * boot. Without this, stuck `media_job` rows would never get cleaned up.
+ */
+const MEDIA_JOB_REAP_INTERVAL_MS = 5 * 60 * 1_000;
+// BullMQ requires a dedicated connection per Worker; give the Queue its own
+// so the scheduler upsert doesn't piggyback on the worker's blocking client.
+const schedulerConnection = createRedisConnection();
+const maintenanceQueue = new Queue(QUEUE_NAMES.maintenance, { connection: schedulerConnection });
+
+async function registerSchedulers(): Promise<void> {
+  try {
+    await maintenanceQueue.upsertJobScheduler(
+      'media-job-reap-scheduler',
+      { every: MEDIA_JOB_REAP_INTERVAL_MS },
+      { name: JOB_NAMES.maintenance.mediaJobReap, data: {} }
+    );
+    log.info('maintenance-scheduler-registered', {
+      name: JOB_NAMES.maintenance.mediaJobReap,
+      everyMs: MEDIA_JOB_REAP_INTERVAL_MS
+    });
+  } catch (err) {
+    log.error('maintenance-scheduler-register-failed', { error: errorMessage(err) });
+  }
+}
+
+void registerSchedulers();
+
 worker.on('ready', () => log.info('maintenance-worker-ready'));
 worker.on('failed', (job, err) =>
   log.error('maintenance-job-failed', { jobName: job?.name, error: errorMessage(err) })
@@ -50,6 +96,8 @@ worker.on('error', (err) => log.error('maintenance-worker-error', { error: error
 const shutdown = async (signal: string) => {
   log.info('maintenance-worker-shutdown', { signal });
   await worker.close();
+  await maintenanceQueue.close();
+  await schedulerConnection.quit();
   await connection.quit();
   process.exit(0);
 };