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 killed formatter grandchild as long to die as it had to start (#12086)

Daniel Hand4d224ebab44
diff --git a/tests/test_formatter_fixed_point.py b/tests/test_formatter_fixed_point.py
index 0f7bdaea07b..3153d78a7fa 100644
--- a/tests/test_formatter_fixed_point.py
+++ b/tests/test_formatter_fixed_point.py
@@ -516,9 +516,12 @@ def test_a_timeout_in_the_polling_loop_kills_the_running_formatters(tmp_path, mo
assert all(proc.poll() is not None for proc in started), "a formatter outlived the timeout"
grandchild = _read_pid(pid_file)
assert grandchild is not None, "the batch never started its child"
- for _ in range(100):
- if not _alive(grandchild):
- break
+ # SIGKILL is sent at once, but the process only dies once it is next scheduled, and a
+ # grandchild still in Python's startup can sit in uninterruptible I/O on a loaded runner.
+ # Five seconds was not always enough under xdist (Repo tests (CPU, rest) on #12060), so wait
+ # as long as the launch above may take. A group kill that missed it still fails, just later.
+ deadline = real_monotonic() + 30
+ while _alive(grandchild) and real_monotonic() < deadline:
real_sleep(0.05)
assert not _alive(grandchild), "the formatter's own child outlived the timeout"