fix: reset task status to queued on intermediate BullMQ retry attempts

backend/src/__tests__/upload.worker.test.ts

@@ -106,11 +106,11 @@ describe('upload worker — retryable errors escape the processor without a DB u
 // it calls updateUploadStatus('failed') so the row no longer stays permanently
 // stuck at 'in_progress'.
 
-describe('upload worker — failed handler skips DB on intermediate BullMQ attempt', () => {
+describe('upload worker — failed handler resets status to queued on intermediate BullMQ attempt', () => {
   it.each([
     ['HashLockError', new HashLockError('lock already held')],
     ['SourceCdnError', new SourceCdnError('cdn network error (ECONNRESET)')],
-  ])('%s: does not touch the DB when BullMQ will retry', async (_name, error) => {
+  ])('%s: resets status to queued when BullMQ will retry', async (_name, error) => {
     const failedHandler = capturedHandlers.get('failed')
     expect(failedHandler).toBeDefined()
 
@@ -119,9 +119,20 @@ describe('upload worker — failed handler skips DB on intermediate BullMQ attem
 
     await failedHandler!(job, error)
 
-    expect(mockUpdateStatus).not.toHaveBeenCalled()
+    expect(mockUpdateStatus).toHaveBeenCalledWith(1, 'queued')
     expect(mockClearToken).not.toHaveBeenCalled()
   })
+
+  it('swallows DB errors when resetting status to queued so the handler never rejects', async () => {
+    mockUpdateStatus.mockRejectedValueOnce(new Error('DB connection lost'))
+
+    const failedHandler = capturedHandlers.get('failed')
+    expect(failedHandler).toBeDefined()
+
+    const job = { ...makeJob(1), attemptsMade: 1, opts: { attempts: 3 } }
+
+    await expect(failedHandler!(job, new SourceCdnError('cdn error'))).resolves.toBeUndefined()
+  })
 })
 
 describe('upload worker — permanent BullMQ failure marks upload as failed in DB', () => {

backend/src/workers/upload.worker.ts

@@ -253,12 +253,20 @@ export function createUploadWorker(redis: Redis, deps?: WorkerDeps): Worker<Uplo
     if (!job) return
     // BullMQ fires 'failed' on every attempt, not just the final one.
     // For errors that rely on BullMQ's built-in retry (not StorageError's custom requeue),
-    // skip DB updates on intermediate attempts — the access token must stay intact for retries.
+    // reset status to queued on intermediate attempts — the access token must stay intact for retries.
     if (!(err instanceof StorageError) && job.attemptsMade < (job.opts.attempts ?? 1)) {
       logger.warn(
         { jobId: job.id, err },
         `[worker] [${uploadId}/${batchId}] job attempt failed, will retry`,
       )
+      try {
+        await uploads.updateUploadStatus(job.data.uploadId, 'queued')
+      } catch (dbErr) {
+        logger.error(
+          { jobId: job.id, err: dbErr },
+          `[worker] [${uploadId}/${batchId}] failed to update db status after job failure`,
+        )
+      }
       return
     }
     logger.error({ jobId: job.id, err }, `[worker] [${uploadId}/${batchId}] job permanently failed`)