fix: preserve live tasks during daemon recovery

This commit is contained in:
chenmingxuan committed 2026-10-08 17:33:48 +08:00
1 parent 079d7b1e28
commit 0355af7833
11 files changed
+327 -17

No files matched your search

+31 -13
View File
@@ -24,6 +24,7 @@ import os
from pathlib import Path
import secrets
import signal
import socket
import sys
from types import FrameType
@@ -421,6 +422,14 @@ queue_goals = state.queue_goals
class ArtemisUvicornServer(uvicorn.Server):
"""Notify application streams before Uvicorn waits for them to close."""
async def startup(self, sockets: list[socket.socket] | None = None) -> None:
# Reload workers start in a fresh interpreter; restore the actual bind
# address there before lifespan publishes runtime metadata.
state.host = self.config.host
state.port = self.config.port
app.state.uvicorn_server = self
await super().startup(sockets=sockets)
@staticmethod
def _task_is_active() -> bool:
"""Check task state without doing process I/O from a signal handler."""
@@ -470,31 +479,40 @@ 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."""
# Reserve a listening socket before publishing metadata or running lifespan.
# Uvicorn normally runs application startup *before* binding its port; a
# duplicate daemon would otherwise recover jobs owned by the real server.
family = socket.AF_INET6 if ":" in host else socket.AF_INET
try:
listener = socket.create_server((host, port), family=family)
except OSError as exc:
print(f"无法启动服务:{host}:{port} 端口不可用({exc})。现有服务和任务保持不变。")
raise SystemExit(1) from exc
listener.set_inheritable(True)
port = listener.getsockname()[1]
state.host = host
state.port = port
write_server_info(port=port, host=host, lifecycle_token=LIFECYCLE_TOKEN)
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,
"apps.admin_console.server:app" if reload else app,
host=host,
port=port,
reload=reload,
timeout_graceful_shutdown=5,
)
if reload:
from uvicorn.supervisors import ChangeReload
server = ArtemisUvicornServer(config)
ChangeReload(config, target=server.run, sockets=[listener]).run()
return
server = ArtemisUvicornServer(config)
app.state.uvicorn_server = server
server.run()
server.run(sockets=[listener])
finally:
listener.close()
app.state.uvicorn_server = None
clear_server_info(port=port, lifecycle_token=LIFECYCLE_TOKEN)