image-sweeper: raise on job failure

ariadne/services/image_sweeper.py

@@ -194,7 +194,15 @@ class ImageSweeperService:
             extra={"event": "image_sweeper_trigger", "job": name},
         )
         if wait:
-            return self._wait_for_completion(name, settings.image_sweeper_wait_timeout_sec)
+            result = self._wait_for_completion(name, settings.image_sweeper_wait_timeout_sec)
+            status = result.get("status")
+            if status != "ok":
+                logger.error(
+                    "image sweeper job incomplete",
+                    extra={"event": "image_sweeper_incomplete", "job": name, "status": status},
+                )
+                raise RuntimeError(f"image sweeper job {name} {status}")
+            return result
         return {"job": name, "status": "queued"}
 
 

tests/test_image_sweeper.py

@@ -46,3 +46,31 @@ def test_image_sweeper_run_wait(monkeypatch) -> None:
     svc = ImageSweeperService()
     result = svc.run(wait=True)
     assert result["status"] == "ok"
+
+
+def test_image_sweeper_run_error(monkeypatch) -> None:
+    dummy_settings = types.SimpleNamespace(
+        image_sweeper_namespace="maintenance",
+        image_sweeper_service_account="node-image-sweeper",
+        image_sweeper_job_ttl_sec=3600,
+        image_sweeper_wait_timeout_sec=30.0,
+    )
+    monkeypatch.setattr(sweeper_module, "settings", dummy_settings)
+
+    def fake_post(path, payload):
+        assert path.endswith("/maintenance/jobs")
+        return {"metadata": {"name": "job-1"}}
+
+    def fake_get(_path):
+        return {"status": {"failed": 1}}
+
+    monkeypatch.setattr(sweeper_module, "post_json", fake_post)
+    monkeypatch.setattr(sweeper_module, "get_json", fake_get)
+
+    svc = ImageSweeperService()
+    try:
+        svc.run(wait=True)
+    except RuntimeError as exc:
+        assert "image sweeper job" in str(exc)
+    else:
+        raise AssertionError("expected image sweeper failure to raise")