Fix: avoid IntegrityError when a retried task republishes with the same task_id (#14096)

src/documents/signals/handlers.py

@@ -1189,13 +1189,18 @@ def before_task_publish_handler(
         trigger_source = _determine_trigger_source(headers)
         owner_id = _extract_owner_id(task_type, task_kwargs)
 
-        PaperlessTask.objects.create(
+        # A retried task is republished with the same task_id, so this fires
+        # again for it; get_or_create keeps the original PENDING record
+        # instead of raising a duplicate-key IntegrityError on the retry.
+        PaperlessTask.objects.get_or_create(
             task_id=task_id,
-            task_type=task_type,
-            trigger_source=trigger_source,
-            status=PaperlessTask.Status.PENDING,
-            input_data=input_data,
-            owner_id=owner_id,
+            defaults={
+                "task_type": task_type,
+                "trigger_source": trigger_source,
+                "status": PaperlessTask.Status.PENDING,
+                "input_data": input_data,
+                "owner_id": owner_id,
+            },
         )
     except Exception:  # pragma: no cover
         logger.exception("Creating PaperlessTask failed")

src/documents/tests/test_task_signals.py

@@ -106,6 +106,17 @@ class TestBeforeTaskPublishHandler:
         assert task.task_type == PaperlessTask.TaskType.TRAIN_CLASSIFIER
         assert task.trigger_source == PaperlessTask.TriggerSource.MANUAL
 
+        # A Celery retry republishes with the same task_id; this must not
+        # raise a duplicate-key IntegrityError, and must leave the original
+        # PENDING record alone.
+        send_publish(
+            "documents.tasks.train_classifier",
+            (),
+            {},
+            headers={"id": task_id},
+        )
+        assert PaperlessTask.objects.filter(task_id=task_id).count() == 1
+
     def test_creates_task_for_sanity_check(self) -> None:
         task_id = send_publish("documents.tasks.sanity_check", (), {})
         task = PaperlessTask.objects.get(task_id=task_id)