fix(test): make queue completion test more robust (#663)

Co-authored-by: Jason Woltje <jason@diversecanvas.com>
Co-committed-by: Jason Woltje <jason@diversecanvas.com>

.woodpecker/ci.yml

@@ -337,3 +337,44 @@ steps:
       - security-trivy-api
       - security-trivy-orchestrator
       - security-trivy-web
+
+  # ─── Deploy to Docker Swarm (main only) ─────────────────────
+
+  # ─── Deploy to Docker Swarm via Portainer (main only) ─────────────────────
+
+  deploy-swarm:
+    image: alpine:3
+    environment:
+      SSH_PRIVATE_KEY:
+        from_secret: ssh_private_key
+      SSH_KNOWN_HOSTS:
+        from_secret: ssh_known_hosts
+      PORTAINER_URL:
+        from_secret: portainer_url
+      PORTAINER_API_KEY:
+        from_secret: portainer_api_key
+    commands:
+      - apk add --no-cache curl openssh-client
+      - |
+        set -e
+        echo "🚀 Deploying to Docker Swarm..."
+
+        # Setup SSH for fallback
+        mkdir -p ~/.ssh
+        echo "$SSH_KNOWN_HOSTS" > ~/.ssh/known_hosts
+        chmod 600 ~/.ssh/known_hosts
+        echo "$SSH_PRIVATE_KEY" > ~/.ssh/id_ed25519
+        chmod 600 ~/.ssh/id_ed25519
+
+        # Force service updates (images are pulled from public registry)
+        ssh -o StrictHostKeyChecking=no localadmin@10.1.1.45 \
+          "docker service update --with-registry-auth --force mosaic-stack-api && \
+           docker service update --with-registry-auth --force mosaic-stack-web && \
+           docker service update --with-registry-auth --force mosaic-stack-orchestrator && \
+           docker service update --with-registry-auth --force mosaic-stack-coordinator && \
+           echo '✅ All services updated'"
+    when:
+      - branch: [main]
+        event: [push, manual, tag]
+    depends_on:
+      - link-packages

apps/coordinator/tests/test_coordinator.py

@@ -601,9 +601,21 @@ class TestCoordinatorIntegration:
         coordinator = Coordinator(queue_manager=queue_manager, poll_interval=0.02)
 
         task = asyncio.create_task(coordinator.start())
-        await asyncio.sleep(0.5)  # Allow time for processing
-        await coordinator.stop()
 
+        # Poll for completion with timeout instead of fixed sleep
+        deadline = asyncio.get_event_loop().time() + 5.0  # 5 second timeout
+        while asyncio.get_event_loop().time() < deadline:
+            all_completed = True
+            for i in range(157, 162):
+                item = queue_manager.get_item(i)
+                if item is None or item.status != QueueItemStatus.COMPLETED:
+                    all_completed = False
+                    break
+            if all_completed:
+                break
+            await asyncio.sleep(0.05)
+
+        await coordinator.stop()
         task.cancel()
         try:
             await task