feat: add multi-device concurrency, server lifecycle CLI, and unified execution stream

- Implement DevicePool and DeviceExecutionLock concurrency modes (per-device FIFO vs global serialization)
- Extend ArtemisClient SDK and CLI with explicit device targeting and concurrency strategy configuration
- Add server lifecycle management (artemis restart, stop, status) with cross-platform PID discovery
- Update MCP server tools to support multi-device dispatch, trace tagging, and per-device execution
- Enhance Showcase UI with step-by-step visual replay, playback speed controls, and device serial badges
- Add comprehensive unit test coverage and edge-case verification scripts
This commit is contained in:
somew1nd committed 2026-08-26 15:53:14 -07:00
1 parent a36a66ce6d
commit c2c90a6378
84 files changed
+5612 -645

No files matched your search

+2
View File
@@ -30,6 +30,8 @@ class ServerState:
self.ipc_server: asyncio.Server | None = None
self.ipc_serve_task: asyncio.Task | None = None
self.ipc_port: int | None = None
self.port: int = 8000
self.host: str = "0.0.0.0"
self.is_shutting_down: bool = False
self.current_process: asyncio.subprocess.Process | None = None
+3 -1
View File
@@ -1148,13 +1148,15 @@
const time = new Date(session.start_time * 1000).toLocaleString();
const titleText = session.initial_goal || 'Unnamed Session';
const sidDisplay = session.session_id || '';
const deviceSerial = session.device_serial || session.device_id || '';
const devicePrefix = deviceSerial ? `${escapeHtml(deviceSerial)} · ` : '';
div.innerHTML = `
<div style="display: flex; justify-content: space-between; align-items: flex-start;">
<div style="flex: 1; min-width: 0;">
<h3 style="font-size: 0.95rem; margin-bottom: 3px; margin-top: 0; white-space: nowrap; overflow: hidden; text-overflow: ellipsis;" title="${titleText}">${titleText}</h3>
<div style="font-size: 0.75rem; font-family: monospace; opacity: 0.85; margin-bottom: 3px; word-break: break-all; color: ${session.session_id === currentSessionId ? '#edf2f7' : '#4a5568'};">ID: ${sidDisplay}</div>
<small style="color: ${session.session_id === currentSessionId ? '#e2e8f0' : '#718096'}; font-size: 0.75rem;">${time}</small>
<small style="color: ${session.session_id === currentSessionId ? '#e2e8f0' : '#718096'}; font-size: 0.75rem;">${devicePrefix}${time}</small>
</div>
<button onclick="deleteSession(event, '${session.session_id}')" style="background: none; border: none; color: ${session.session_id === currentSessionId ? '#fff' : '#e53e3e'}; cursor: pointer; font-size: 1.1rem; padding: 0 0 0 8px; line-height: 1; flex-shrink: 0;" title="Delete Session">🗑️</button>
</div>
+30 -2
View File
@@ -12,6 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import json
import time
from uuid import UUID
from fastapi import APIRouter, HTTPException
@@ -49,12 +50,39 @@ async def list_sessions():
unresolved_profiles = []
result = []
try:
from artemis.runtime import DeviceExecutionLock
active_owners = DeviceExecutionLock.get_active_owners()
active_owner_sids = {
str(owner.session_id)
for owner in active_owners.values()
if owner.session_id
}
except Exception:
active_owner_sids = set()
for row_dict in rows:
s_id = str(row_dict.get("session_id"))
recording = latest_recordings.get(s_id)
d_info_raw = row_dict.get("device_info")
device_id = None
if d_info_raw:
try:
d_info = json.loads(d_info_raw) if isinstance(d_info_raw, str) else d_info_raw
if isinstance(d_info, dict):
device_id = d_info.get("device_id") or d_info.get("device_serial")
except Exception:
pass
if not device_id and recording:
device_id = recording.get("device_id")
row_dict["device_id"] = device_id
row_dict["device_serial"] = device_id
if row_dict.get("status") == "running":
is_active = s_id in state.active_connections or (
state.is_running and s_id == str(state.active_session_id)
is_active = (
s_id in active_owner_sids
or s_id in state.active_connections
or (state.is_running and s_id == str(state.active_session_id))
)
worker_is_alive = session_repo.process_is_alive(row_dict.get("pid"))
if not is_active and not worker_is_alive:
+93
View File
@@ -360,3 +360,96 @@ async def get_model_config_and_env():
"env_filename": ".env",
"env_vars": env_vars,
}
@router.get("/server-status")
async def get_server_runtime_status():
"""Retrieve runtime status, PID, port, and uptime of the Artemis server."""
import os
from artemis.runtime.server_lifecycle import get_server_status
try:
from apps.admin_console.core.state import state
except ImportError:
from admin_console.core.state import state
port = getattr(state, "port", 8000)
status = get_server_status(port=port)
status["current_pid"] = os.getpid()
return status
@router.post("/restart")
async def restart_server_endpoint():
"""Request a graceful restart of the Artemis server from thin clients/UI."""
import asyncio
import os
import sys
import threading
try:
from apps.admin_console.core.state import state
except ImportError:
from admin_console.core.state import state
port = getattr(state, "port", 8000)
current_pid = os.getpid()
def _restart_worker():
import time
time.sleep(0.6)
if sys.platform != "win32":
try:
os.execv(sys.executable, [sys.executable] + sys.argv)
except Exception:
import subprocess
subprocess.Popen([sys.executable] + sys.argv)
os._exit(0)
else:
import subprocess
subprocess.Popen([sys.executable] + sys.argv)
os._exit(0)
threading.Thread(target=_restart_worker, daemon=True).start()
return {
"status": "restarting",
"message": "Artemis server is restarting. Client reconnection should occur in 2-3 seconds.",
"previous_pid": current_pid,
"port": port,
}
@router.post("/shutdown")
async def shutdown_server_endpoint():
"""Request a graceful shutdown of the Artemis server."""
import os
import signal
import threading
try:
from apps.admin_console.core.state import state
except ImportError:
from admin_console.core.state import state
current_pid = os.getpid()
def _shutdown_worker():
import time
time.sleep(0.5)
try:
os.kill(current_pid, signal.SIGTERM)
except Exception:
os._exit(0)
threading.Thread(target=_shutdown_worker, daemon=True).start()
return {
"status": "shutting_down",
"message": "Artemis server is shutting down.",
"pid": current_pid,
}
+46 -3
View File
@@ -18,7 +18,7 @@ import json
from fastapi import APIRouter, HTTPException
from fastapi.responses import StreamingResponse
from artemis.core.diagnostics import readiness_engine
from artemis.runtime import DeviceExecutionLock
from artemis.runtime import DeviceExecutionLock, device_pool
try:
from admin_console.core.state import state
@@ -99,9 +99,17 @@ async def run_task(request: RunRequest):
enable_outputter=request.enable_outputter,
locked_app_package=request.locked_app_package,
app_path=request.app_path,
device_serial=request.device_serial,
)
@router.get("/api/devices")
async def list_devices():
"""List all connected Android devices with their busy / idle status."""
devices = await device_pool.list_devices_async()
return {"devices": [d.to_dict() for d in devices]}
@router.post("/api/stop")
async def stop_task(all: bool = False):
stopped = task_queue_service.stop_tasks(clear_all=all)
@@ -138,10 +146,24 @@ async def get_status():
None,
)
active_owners = DeviceExecutionLock.get_active_owners()
active_tasks = [
{
"device_id": owner.device_id,
"session_id": owner.session_id,
"goal": owner.description,
"pid": owner.pid,
"ingress": owner.ingress,
"acquired_at": owner.acquired_at,
}
for owner in active_owners.values()
]
running_sid = (
(global_owner.session_id if global_owner else None)
or state.active_session_id
or (running_task.get("session_id") if running_task else None)
or (active_tasks[0]["session_id"] if active_tasks and active_tasks[0].get("session_id") else None)
or (session_repo.get_running_session_id() if is_running else None)
)
owner_connection = (
@@ -152,6 +174,7 @@ async def get_status():
or (global_owner.description if global_owner else None)
or state.current_goal
or (running_task.get("goal") if running_task else None)
or (active_tasks[0]["goal"] if active_tasks else None)
)
active_profile = owner_connection.get("profile") or state.current_profile or (
@@ -168,7 +191,23 @@ async def get_status():
)
model_info = model_service.get_active_model_info(active_profile)
queue_data = state.queue_tasks
# Unified Global Queue: merge web tasks and external SDK/CLI device queue tickets
global_queued = DeviceExecutionLock.get_queued_tasks()
seen_ids = set()
queue_data: list[dict[str, Any]] = []
for item in state.queue_tasks:
sid = item.get("session_id")
if sid:
seen_ids.add(str(sid))
queue_data.append(item)
for g_item in global_queued:
sid = str(g_item.get("session_id"))
if sid not in seen_ids:
seen_ids.add(sid)
queue_data.append(g_item)
if is_running:
is_paused = state.is_paused
@@ -187,6 +226,7 @@ async def get_status():
"background_tasks": bg_tasks,
"queue": queue_data,
"model_info": model_info,
"active_tasks": active_tasks,
}
if latest_session_id and str(latest_session_id) in state.active_connections:
@@ -203,6 +243,8 @@ async def get_status():
"background_tasks": bg_tasks,
"queue": queue_data,
"model_info": conn_model_info,
"active_tasks": active_tasks,
"ipc_port": state.ipc_port,
}
return {
@@ -210,7 +252,8 @@ async def get_status():
"session_id": latest_session_id,
"background_tasks": bg_tasks,
"queue": queue_data,
"model_info": model_info,
"active_tasks": active_tasks,
"ipc_port": state.ipc_port,
}
@@ -23,6 +23,8 @@ class RunRequest(BaseModel):
enable_outputter: bool | None = None
locked_app_package: str | None = None
app_path: str | None = None
device_serial: str | None = None
class ReplayRequest(BaseModel):
+23 -12
View File
@@ -49,8 +49,10 @@ import uvicorn
from artemis.runtime import (
DeviceExecutionLock,
clear_server_info,
shutdown_awake_service,
start_awake_service,
write_server_info,
)
try:
@@ -108,6 +110,7 @@ async def on_startup():
"""Startup lifecycle hooks."""
state.is_shutting_down = False
state.shutdown_event.clear()
write_server_info(port=getattr(state, "port", 8000), host=getattr(state, "host", "0.0.0.0"))
await asyncio.to_thread(start_awake_service)
cleaned_device_locks = DeviceExecutionLock.cleanup_stale_locks()
if cleaned_device_locks:
@@ -160,6 +163,7 @@ async def on_shutdown():
await ipc_service.stop_server()
state.ipc_subscribers.clear()
await asyncio.to_thread(shutdown_awake_service)
clear_server_info()
# Mount modular routers
@@ -382,23 +386,30 @@ class ArtemisUvicornServer(uvicorn.Server):
def run_ui_server(host: str, port: int, reload: bool = False) -> None:
"""Run the UI server with bounded, signal-aware graceful shutdown."""
if reload:
uvicorn.run(
state.host = host
state.port = port
write_server_info(port=port, host=host)
try:
if reload:
uvicorn.run(
"apps.admin_console.server:app",
host=host,
port=port,
reload=True,
timeout_graceful_shutdown=5,
)
return
config = uvicorn.Config(
app,
host=host,
port=port,
reload=True,
timeout_graceful_shutdown=5,
)
return
config = uvicorn.Config(
app,
host=host,
port=port,
timeout_graceful_shutdown=5,
)
ArtemisUvicornServer(config).run()
ArtemisUvicornServer(config).run()
finally:
clear_server_info()
def main():
@@ -334,12 +334,12 @@ PRESET_TASK_CATALOG: list[TaskPreset] = [
required_packages=["com.google.android.apps.photos"],
priority=82,
),
# 9. WeChat (微信)
# 9. WeChat
TaskPreset(
id="wechat_browse",
title="Check WeChat Messages",
description="Open WeChat and view top recent chat conversations",
goal="Open WeChat (微信) and view the top recent chat messages.",
goal="Open WeChat and view the top recent chat messages.",
profile="flash",
category="flash",
tag="WeChat",
@@ -351,7 +351,7 @@ PRESET_TASK_CATALOG: list[TaskPreset] = [
id="pro_wechat_to_calendar",
title="WeChat Notice to Calendar",
description="Extract meeting notice from WeChat chat and add to Calendar",
goal="Open WeChat (微信), locate the latest meeting announcement or event message in the top chat, extract the time and topic, then open Calendar and schedule an event.",
goal="Open WeChat, locate the latest meeting announcement or event message in the top chat, extract the time and topic, then open Calendar and schedule an event.",
profile="pro",
category="cross_app",
tag="WeChat + Calendar",
@@ -367,12 +367,12 @@ PRESET_TASK_CATALOG: list[TaskPreset] = [
required_packages=["com.tencent.mm"],
priority=93,
),
# 10. Xiaohongshu (小红书)
# 10. Xiaohongshu
TaskPreset(
id="xhs_coffee_guide",
title="RED Cafe Guide Search",
description="Search trending specialty cafe reviews on Xiaohongshu",
goal="Open Xiaohongshu (小红书), search for top-rated specialty coffee shops, and view the top post.",
goal="Open Xiaohongshu, search for top-rated specialty coffee shops, and view the top post.",
profile="flash",
category="flash",
tag="Xiaohongshu",
@@ -384,12 +384,12 @@ PRESET_TASK_CATALOG: list[TaskPreset] = [
required_packages=["com.xingin.xhs"],
priority=87,
),
# 11. Meituan / Dianping (美团 / 大众点评)
# 11. Meituan / Dianping
TaskPreset(
id="meituan_ramen_search",
title="Meituan Food Search",
description="Search top-rated Ramen nearby on Meituan or Dianping",
goal="Open Meituan (美团) or Dianping (大众点评), search for top-rated Ramen nearby, and view top restaurant rating.",
goal="Open Meituan or Dianping, search for top-rated Ramen nearby, and view top restaurant rating.",
profile="flash",
category="flash",
tag="Meituan",
@@ -401,12 +401,12 @@ PRESET_TASK_CATALOG: list[TaskPreset] = [
required_packages=["com.sankuai.meituan", "com.dianping.v1"],
priority=86,
),
# 12. Bilibili (哔哩哔哩)
# 12. Bilibili
TaskPreset(
id="bilibili_stream",
title="Bilibili Tech Video",
description="Search and play an AI Agent tutorial video on Bilibili",
goal='Open Bilibili (哔哩哔哩), search for "AI Agent 架构实战", and play the top matching video.',
goal='Open Bilibili, search for "AI Agent Architecture", and play the top matching video.',
profile="flash",
category="flash",
tag="Bilibili",
@@ -241,6 +241,11 @@ class TaskQueueService:
async def queue_worker(cls):
"""Persistent, self-healing background worker loop processing tasks in FIFO order."""
print("[QueueWorker] Background worker loop initialized and running.")
try:
DeviceExecutionLock.cleanup_stale_locks()
except Exception as exc:
print(f"[QueueWorker] Initial stale lock cleanup notice: {exc}")
while True:
task_item = None
sess_id = None
@@ -289,6 +294,7 @@ class TaskQueueService:
"session_id": sess_id,
"initial_goal": goal,
"profile": profile,
"device_serial": task_item.get("device_serial"),
},
)
@@ -322,6 +328,8 @@ class TaskQueueService:
"--test-name",
test_name,
]
if sess_id:
cmd.extend(["--session-id", str(sess_id)])
if expected_output:
cmd.extend(["--output-description", str(expected_output)])
if enable_outputter is not None:
@@ -330,9 +338,13 @@ class TaskQueueService:
cmd.extend(["--locked-app", str(locked_app)])
if app_path:
cmd.extend(["--app-path", str(app_path)])
device_serial = task_item.get("device_serial")
if device_serial:
cmd.extend(["--device-serial", str(device_serial)])
env["ADB_DEVICE_SERIAL"] = str(device_serial)
print(
f"[QueueWorker] Starting task [{sess_id}]: '{goal}' (profile: {profile}, outputter: {bool(expected_output or enable_outputter)})"
f"[QueueWorker] Starting task [{sess_id}]: '{goal}' (profile: {profile}, device: {device_serial or 'auto'}, outputter: {bool(expected_output or enable_outputter)})"
)
proc = await asyncio.create_subprocess_exec(
*cmd,
@@ -349,6 +361,7 @@ class TaskQueueService:
str(queue_ticket),
proc.pid,
description=f"frontend task: {goal[:120]}",
device_id=device_serial or "pending",
session_id=str(sess_id) if sess_id else None,
ingress="frontend",
)
@@ -453,6 +466,7 @@ class TaskQueueService:
enable_outputter: bool | None = None,
locked_app_package: str | None = None,
app_path: str | None = None,
device_serial: str | None = None,
) -> dict[str, Any]:
"""Enqueues one or more goals and wakes up the background worker."""
cls.ensure_worker_running()
@@ -461,8 +475,18 @@ class TaskQueueService:
now = time.time()
for i, goal in enumerate(goals):
sess_id = str(uuid.uuid4())
assigned_serial = device_serial
if not assigned_serial:
try:
from artemis.runtime import device_pool
assigned_serial = device_pool.select_device()
except Exception:
assigned_serial = None
queue_ticket = DeviceExecutionLock.reserve(
description=f"frontend task: {goal[:120]}",
device_id=assigned_serial or "pending",
session_id=sess_id,
ingress="frontend",
)
@@ -474,6 +498,7 @@ class TaskQueueService:
"enable_outputter": enable_outputter,
"locked_app_package": locked_app_package,
"app_path": app_path,
"device_serial": assigned_serial,
"status": "pending",
"queue_ticket": queue_ticket,
"created_at": now + i * 0.001,