mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-07-09 13:21:42 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 6

fix(computer-use): increase cua-driver session startup timeout from 15s to 30s

Browse source 
On Windows, the cua-driver MCP session initialization can exceed the 15s
timeout due to manifest subprocess discovery + MCP transport setup.
This makes the computer_use tool permanently unavailable even though
hermes computer-use doctor and hermes mcp test both pass.

- Increase _ready_event.wait timeout from 15s to 30s
- Add startup timing instrumentation (manifest + mcp_init durations)
- Log timing at INFO level for diagnosability

Fixes #57025
This commit is contained in:
liuhao1024 2026-07-02 19:55:33 +08:00 • committed by Teknium
parent d8b51269ca
commit d537d29a6f
1 changed files with 13 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

15
tools/computer_use/cua_backend.py
View file

@ -567,6 +567,7 @@ class _CuaDriverSession:
different task than it was entered in" warning emitted by the
previous _aenter/_aexit split.
"""
import time as _time
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
from tools.environments.local import _sanitize_subprocess_env
@ -574,6 +575,7 @@ class _CuaDriverSession:
# Build the shutdown event on the loop's thread so the asyncio
# primitive belongs to the correct loop.
self._shutdown_event = asyncio.Event()
_t0 = _time.monotonic()
try:
if not cua_driver_binary_available():
@ -583,6 +585,7 @@ class _CuaDriverSession:
# the MCP server, instead of hardcoding ["mcp"]. Falls back
# transparently for older drivers / any discovery failure.
command, args = _resolve_mcp_invocation(_CUA_DRIVER_CMD)
_t_manifest = _time.monotonic()
params = StdioServerParameters(
command=command,
args=args,
@ -594,12 +597,20 @@ class _CuaDriverSession:
async with stdio_client(params) as (read, write):
async with ClientSession(read, write) as session:
await session.initialize()
_t_init = _time.monotonic()
# Populate capabilities + capability_version BEFORE
# exposing the session to callers, so the first
# tool call already sees them.
await self._populate_capabilities(session)
self._session = session
self._ready_event.set()
logger.info(
"cua-driver session ready in %.1fs "
"(manifest=%.1fs, mcp_init=%.1fs)",
_time.monotonic() - _t0,
_t_manifest - _t0,
_t_init - _t_manifest,
)
# Hold the contexts open until stop() / restart asks
# us to wind down. Tool calls run as their own tasks
# on the same loop and touch self._session directly.
@ -677,10 +688,10 @@ class _CuaDriverSession:
self._lifecycle_future = asyncio.run_coroutine_threadsafe(
self._lifecycle_coro(), loop
)
if not self._ready_event.wait(timeout=15.0):
if not self._ready_event.wait(timeout=30.0):
# Best-effort: signal shutdown if the future is still alive.
self._signal_shutdown_locked()
raise RuntimeError("cua-driver session never reached ready (timeout 15s)")
raise RuntimeError("cua-driver session never reached ready (timeout 30s)")
# If setup failed, the lifecycle coroutine set _setup_error
# before setting _ready_event. Re-raise it on the caller's thread.
if self._setup_error is not None: