fix(computer_use): stop the cua-driver child on exit

CuaDriverBackend caches a long-lived cua-driver subprocess for the life of
the Hermes process, and stop() was never called from anywhere — the driver
outlived the session that spawned it. #69903 stopped the orphan from pegging
a core by disabling the cursor overlay, but left the process behind; this is
item 3 of #28152 ("Hermes does not keep the driver alive after tool
completion").

Register an atexit hook, mirroring browser_tool's
atexit.register(_emergency_cleanup_all_sessions). atexit only, no signal
handlers, for the prompt_toolkit reason documented there. reset_backend_for_tests
now reuses the same teardown instead of repeating it.
This commit is contained in:
Brooklyn Nicholson 2026-07-26 03:38:23 -05:00
parent d9f1043c33
commit 4e49af94be
2 changed files with 107 additions and 8 deletions

View file

@ -0,0 +1,75 @@
"""Tests for cua-driver subprocess teardown on interpreter exit.
``CuaDriverBackend`` spawns a long-lived ``cua-driver`` child process and
caches it for the life of the Hermes process. Nothing ever tore it down, so
the driver outlived the session that started it (#28152 item 3 — "Hermes does
not keep the driver alive after tool completion"). #69903 fixed the *overlay*
redraw loop that made the lingering process burn CPU, but not the lingering
process itself.
``tools/computer_use/tool.py`` registers an ``atexit`` hook that stops the
cached backend, mirroring ``browser_tool``'s
``atexit.register(_emergency_cleanup_all_sessions)``.
These assert the behavior contract the hook is registered, it stops a live
backend, it is a no-op when nothing was ever started, and it never raises out
of ``atexit`` not a snapshot of the module's source.
"""
import atexit
from unittest.mock import MagicMock, patch
from tools.computer_use import tool as cu_tool
class TestAtexitTeardown:
def test_shutdown_stops_a_live_backend(self):
"""A cached backend is stopped when the interpreter exits."""
fake = MagicMock()
with patch.object(cu_tool, "_backend", fake):
cu_tool._shutdown_backend_atexit()
fake.stop.assert_called_once()
def test_shutdown_clears_the_cached_backend(self):
"""After teardown the module cache is empty, so a later call re-spawns."""
fake = MagicMock()
with patch.object(cu_tool, "_backend", fake):
cu_tool._shutdown_backend_atexit()
assert cu_tool._backend is None
def test_shutdown_is_a_noop_when_never_started(self):
"""No backend was ever created => nothing to stop, no error."""
with patch.object(cu_tool, "_backend", None):
cu_tool._shutdown_backend_atexit() # must not raise
assert cu_tool._backend is None
def test_shutdown_swallows_backend_errors(self):
"""A failing stop() must not raise out of an atexit hook.
Exceptions escaping atexit print a traceback on every exit and can
mask the real exit status.
"""
fake = MagicMock()
fake.stop.side_effect = RuntimeError("driver already dead")
with patch.object(cu_tool, "_backend", fake):
cu_tool._shutdown_backend_atexit() # must not raise
assert cu_tool._backend is None
def test_hook_is_registered_with_atexit(self):
"""Importing the tool module registers the teardown hook.
Verified by unregistering and re-registering: atexit.unregister only
removes a function that was actually registered, so a successful
round-trip proves the import-time registration happened.
"""
atexit.unregister(cu_tool._shutdown_backend_atexit)
try:
with patch.object(cu_tool, "_backend", MagicMock()) as fake:
# Re-register and fire the full atexit chain the way the
# interpreter would, then confirm our hook ran.
atexit.register(cu_tool._shutdown_backend_atexit)
atexit._run_exitfuncs()
fake.stop.assert_called_once()
finally:
atexit.unregister(cu_tool._shutdown_backend_atexit)
atexit.register(cu_tool._shutdown_backend_atexit)

View file

@ -38,6 +38,7 @@ For captures / actions with `capture_after=True`:
from __future__ import annotations
import atexit
import base64
import json
import logging
@ -177,16 +178,39 @@ def _get_backend() -> ComputerUseBackend:
return _backend
def _shutdown_backend_atexit() -> None:
"""Stop the cached backend so the cua-driver child doesn't outlive us.
The backend is cached per-process and holds a long-lived ``cua-driver``
subprocess, so without this the driver survives the Hermes process that
spawned it (#28152 item 3). #69903 kept the orphan from burning a core by
disabling the cursor overlay; the process itself still lingered.
Mirrors ``browser_tool``'s ``atexit.register(_emergency_cleanup_all_sessions)``
same spawn-and-drive-a-subprocess shape. atexit only, no signal handlers:
a ``SystemExit`` raised from a prompt_toolkit key binding corrupts its
coroutine state and makes the process unkillable. Never raises, since an
exception escaping atexit prints a traceback on every exit.
"""
global _backend
# Drop the lock before stop() — teardown budgets 5s and shouldn't block
# an unrelated caller waiting to spawn.
with _backend_lock:
backend, _backend = _backend, None
if backend is None:
return
try:
backend.stop()
except Exception as e:
logger.debug("cua-driver atexit teardown failed: %s", e)
atexit.register(_shutdown_backend_atexit)
def reset_backend_for_tests() -> None: # pragma: no cover
"""Test helper — tear down the cached backend and per-session state."""
global _backend
with _backend_lock:
if _backend is not None:
try:
_backend.stop()
except Exception:
pass
_backend = None
_shutdown_backend_atexit()
_AUX_VISION_ROUTE_CACHE.clear()
with _approval_lock:
_session_auto_approve.clear()