Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Close the WAL keepers a backend test opened before the next one runs (#11633)
Daniel Han3fec461cf141
diff --git a/studio/backend/tests/conftest.py b/studio/backend/tests/conftest.pyindex dd5dce81124..b94a3b504f1 100644--- a/studio/backend/tests/conftest.py+++ b/studio/backend/tests/conftest.py@@ -321,6 +321,31 @@ def _isolate_generation_state():active_generations.reset_for_tests()[email protected](autouse = True)+def _isolate_wal_keepers():+ """Close the WAL keepers a test opened, so the next test starts without them.++ ``storage.studio_db`` holds one keeper connection per database in a module global, and+ ``get_connection`` opens one on its own for any managed account, which is the correct+ behaviour in a server and the reason nothing else closes them in a test. So every test that+ touched a managed account's studio.db, tests/multi_account/test_alice_bob_matrix.py among+ them, handed its keepers to whatever ran next in the same xdist worker, and+ test_wal_keeper_declines_when_the_filesystem_refused_wal's ``assert not _wal_keepers``+ failed with somebody else's account databases. Reproduced by running the matrix ahead of it.++ Only what the test itself added is closed, and through ``close_wal_keeper_for`` so the+ close listeners run exactly as they would in the product.+ """+ from storage import studio_db++ before = set(studio_db._wal_keepers)+ unsupported = set(studio_db._wal_unsupported)+ yield+ for path in set(studio_db._wal_keepers) - before:+ studio_db.close_wal_keeper_for(path)+ studio_db._wal_unsupported.intersection_update(unsupported)++@pytest.fixture(autouse = True)def _isolate_audio_gallery(monkeypatch, tmp_path):"""Keep generated-clip persistence out of the developer's real gallery.