add heavy loggin

core/api/detect_sources.py

@@ -168,7 +168,7 @@ def run_pipeline(req: RunRequest):
     from detect.state import DetectState
 
     local_path = _resolve_video_path(req.video_path)
-    job_id = str(uuid.uuid4())[:8]
+    job_id = str(uuid.uuid4())
 
     if req.skip_vlm:
         os.environ["SKIP_VLM"] = "1"
@@ -200,8 +200,6 @@ def run_pipeline(req: RunRequest):
         source_asset_id=req.source_asset_id,
     )
 
-    import traceback
-
     from detect.graph import PipelineCancelled, set_cancel_check, clear_cancel_check
 
     set_cancel_check(job_id, lambda: job_id in _cancelled_jobs)
@@ -218,9 +216,17 @@ def run_pipeline(req: RunRequest):
             emit.job_complete(job_id, {"status": "cancelled"})
         except Exception as e:
             logger.exception("Pipeline run %s failed: %s", job_id, e)
-            tb = traceback.format_exc()
+            # Mark the current/last stage as error in the graph
+            from detect.graph import _node_states, NODES
+            if job_id in _node_states:
+                states = _node_states[job_id]
+                for node in reversed(NODES):
+                    if states.get(node) in ("running", "done"):
+                        states[node] = "error"
+                        break
+                nodes = [{"id": n, "status": states[n]} for n in NODES]
+                emit.graph_update(job_id, nodes)
             emit.log(job_id, "Pipeline", "ERROR", str(e))
-            emit.log(job_id, "Pipeline", "DEBUG", tb)
             emit.job_complete(job_id, {"status": "failed", "error": str(e)})
         finally:
             _running_jobs.pop(job_id, None)

core/api/detect_sse.py

@@ -26,16 +26,16 @@ router = APIRouter(prefix="/detect", tags=["detect"])
 
 async def _event_generator(job_id: str) -> AsyncGenerator[str, None]:
     cursor = 0
-    timeout = time.monotonic() + 3600  # 1 hour max (detection jobs are long)
+    timeout = time.monotonic() + 3600  # 1 hour max
 
     while time.monotonic() < timeout:
         events, cursor = poll_events(job_id, cursor, prefix=DETECT_EVENTS_PREFIX)
 
         if not events:
-            yield f"event: waiting\ndata: {json.dumps({'job_id': job_id})}\n\n"
-            await asyncio.sleep(0.1)
+            await asyncio.sleep(0.2)
             continue
 
+        is_terminal = False
         for data in events:
             event_type = data.pop("event", "update")
             payload = {**data, "job_id": job_id}
@@ -43,8 +43,15 @@ async def _event_generator(job_id: str) -> AsyncGenerator[str, None]:
             yield f"event: {event_type}\ndata: {json.dumps(payload)}\n\n"
 
             if event_type in TERMINAL_EVENTS:
-                yield f"event: done\ndata: {json.dumps({'job_id': job_id})}\n\n"
-                return
+                is_terminal = True
+
+        if is_terminal:
+            yield f"event: done\ndata: {json.dumps({'job_id': job_id})}\n\n"
+            # Don't return — keep connection alive so EventSource doesn't reconnect.
+            # Just idle until the client disconnects or timeout.
+            while time.monotonic() < timeout:
+                await asyncio.sleep(5)
+            return
 
         await asyncio.sleep(0.05)