services/mam-api/src/routes/jobs.js

@@ -207,7 +207,13 @@ router.post('/:id/retry', async (req, res, next) => {
   }
 });
 
-// ── DELETE /:id - Remove a job ────────────────────────────────────────────────
+// ── DELETE /:id - Remove a job (also handles cancel for active jobs) ─────────
+// BullMQ refuses job.remove() while a job is in the 'active' state. Before this
+// fix the route caught that error and fell through to a misleading 404, so
+// operators couldn't kill a stalled-active job from the UI. Now we detect the
+// active state explicitly: moveToFailed with the magic '0' token bypasses the
+// per-worker lock check and transitions active → failed (freeing the queue's
+// concurrency slot), then remove() drops the row.
 router.delete('/:id', async (req, res, next) => {
   try {
     const { id } = req.params;
@@ -215,16 +221,44 @@ router.delete('/:id', async (req, res, next) => {
     const qType  = colonIdx > -1 ? id.slice(0, colonIdx) : null;
     const bullId = colonIdx > -1 ? id.slice(colonIdx + 1) : id;
 
+    let lastErr = null;
     for (const { queue, type } of QUEUES) {
       if (qType && type !== qType) continue;
+      let job;
       try {
-        const job = await queue.getJob(bullId);
-        if (job) {
-          await job.remove();
-          return res.json({ success: true });
+        job = await queue.getJob(bullId);
+      } catch (err) {
+        // Queue-level lookup error: remember it so we don't mask it with 404.
+        lastErr = err;
+        continue;
+      }
+      if (!job) continue;
+
+      const state = await job.getState();
+      if (state === 'active') {
+        // Token '0' tells BullMQ to skip the worker-lock check — necessary
+        // because the operator-side cancel doesn't hold the worker's lock.
+        try {
+          await job.moveToFailed(new Error('Cancelled by operator'), '0', false);
+        } catch (err) {
+          // Lock owned by a still-living worker; fall back to discard + remove
+          // so at least the result is thrown away and the row is gone.
+          try { await job.discard(); } catch (_) {}
         }
-      } catch { /* try next queue */ }
+      }
+      try {
+        await job.remove();
+      } catch (err) {
+        // Last-resort obliteration of the job row via raw Redis. This is
+        // the path stalled jobs hit when moveToFailed couldn't transition
+        // them either.
+        const client = await queue.client;
+        const prefix = queue.toKey(bullId);
+        await client.del(prefix);
+      }
+      return res.json({ success: true, cancelled: state === 'active' });
     }
+    if (lastErr) return next(lastErr);
     res.status(404).json({ error: 'Job not found' });
   } catch (err) {
     next(err);

services/worker/src/index.js

@@ -48,10 +48,20 @@ const createWorker = (queueName, handler, overrides = {}) => {
   return worker;
 };
 
+// Per-queue concurrency. Defaults to 1, which serialises every job in a
+// queue — meaning a single stalled job blocks every other one. We want
+// thumbnails (cheap, parallel-safe) to run several at a time so a slow
+// outlier doesn't back the rest of the catalog up. Proxy + conform are
+// heavier (ffmpeg transcode) so we keep them lower to avoid trashing
+// the box; tune via env if a node has more headroom.
+const PROXY_CONCURRENCY     = parseInt(process.env.PROXY_CONCURRENCY     || '2', 10);
+const THUMBNAIL_CONCURRENCY = parseInt(process.env.THUMBNAIL_CONCURRENCY || '4', 10);
+const CONFORM_CONCURRENCY   = parseInt(process.env.CONFORM_CONCURRENCY   || '1', 10);
+
 const workers = [
-  createWorker('proxy', proxyWorker),
-  createWorker('thumbnail', thumbnailWorker),
-  createWorker('conform', conformWorker),
+  createWorker('proxy',     proxyWorker,     { concurrency: PROXY_CONCURRENCY }),
+  createWorker('thumbnail', thumbnailWorker, { concurrency: THUMBNAIL_CONCURRENCY }),
+  createWorker('conform',   conformWorker,   { concurrency: CONFORM_CONCURRENCY }),
   // YouTube imports: keep concurrency at 1 so we don't burn through rate
   // limits when several jobs land back-to-back. Lock window is longer than
   // the default because a long video download can run for minutes.
@@ -62,6 +72,8 @@ const workers = [
   }),
 ];
 
+console.log(`Concurrency: proxy=${PROXY_CONCURRENCY} thumbnail=${THUMBNAIL_CONCURRENCY} conform=${CONFORM_CONCURRENCY} import=1`);
+
 startPromotionWorker();
 
 console.log('Wild Dragon Worker Service started');