Skip to content

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.

Give the descendant-work health test room for its second interpreter to start (#11670)

Daniel Hanbeab871dc1d6
diff --git a/studio/backend/tests/test_llama_cpp_wait_for_health.py b/studio/backend/tests/test_llama_cpp_wait_for_health.py
index 797de38851d..f8fb15c5237 100644
--- a/studio/backend/tests/test_llama_cpp_wait_for_health.py
+++ b/studio/backend/tests/test_llama_cpp_wait_for_health.py
@@ -1703,7 +1703,12 @@ class TestHealthWaitMeasuresStalls:
"subprocess.Popen([sys.executable, '-c', sys.argv[1], '3.0'])\n"
"time.sleep(60)\n"
)
- b, ok, elapsed = self._wait_on_child(monkeypatch, [shim, self._WORKER], healthy_after = 2.5)
+ # A second interpreter has to start before the descendant does any work, and on a loaded runner that
+ # alone outlasted the 0.6s default: the wait gave up on a load that was about to make progress. 1.5s
+ # still sits well under healthy_after, so only descendant work can carry the wait to 2.5s.
+ b, ok, elapsed = self._wait_on_child(
+ monkeypatch, [shim, self._WORKER], healthy_after = 2.5, timeout = 1.5
+ )
assert ok is True
assert elapsed >= 2.5