Fix shutdown timeouts (#695)

Co-authored-by: AAGaming <aagaming@riseup.net>
This commit is contained in:
Sims
2024-09-01 20:15:49 +02:00
committed by GitHub
parent 4842a599e0
commit 016ed6e998
9 changed files with 70 additions and 44 deletions
+6 -1
View File
@@ -104,10 +104,15 @@ class Loader:
async def enable_reload_wait(self):
if self.live_reload:
await sleep(10)
if self.watcher:
if self.watcher and self.live_reload:
self.logger.info("Hot reload enabled")
self.watcher.disabled = False
async def disable_reload(self):
if self.watcher:
self.watcher.disabled = True
self.live_reload = False
async def handle_frontend_assets(self, request: web.Request):
file = Path(__file__).parent.joinpath("static").joinpath(request.match_info["path"])
return web.FileResponse(file, headers={"Cache-Control": "no-cache"})
@@ -7,22 +7,24 @@ from .localplatform import ON_WINDOWS
BUFFER_LIMIT = 2 ** 20 # 1 MiB
class UnixSocket:
def __init__(self, on_new_message: Callable[[str], Coroutine[Any, Any, Any]]):
def __init__(self):
'''
on_new_message takes 1 string argument.
It's return value gets used, if not None, to write data to the socket.
Method should be async
'''
self.socket_addr = f"/tmp/plugin_socket_{time.time()}"
self.on_new_message = on_new_message
self.on_new_message = None
self.socket = None
self.reader = None
self.writer = None
self.server_writer = None
self.open_lock = asyncio.Lock()
self.active = True
async def setup_server(self):
async def setup_server(self, on_new_message: Callable[[str], Coroutine[Any, Any, Any]]):
try:
self.on_new_message = on_new_message
self.socket = await asyncio.start_unix_server(self._listen_for_method_call, path=self.socket_addr, limit=BUFFER_LIMIT)
except asyncio.CancelledError:
await self.close_socket_connection()
@@ -58,6 +60,8 @@ class UnixSocket:
if self.socket:
self.socket.close()
await self.socket.wait_closed()
self.active = False
async def read_single_line(self) -> str|None:
reader, _ = await self.get_socket_connection()
@@ -81,7 +85,7 @@ class UnixSocket:
async def _read_single_line(self, reader: asyncio.StreamReader) -> str:
line = bytearray()
while True:
while self.active:
try:
line.extend(await reader.readuntil())
except asyncio.LimitOverrunError:
@@ -91,7 +95,7 @@ class UnixSocket:
line.extend(err.partial)
break
except asyncio.CancelledError:
break
raise
else:
break
@@ -111,7 +115,7 @@ class UnixSocket:
async def _listen_for_method_call(self, reader: asyncio.StreamReader, writer: asyncio.StreamWriter):
self.server_writer = writer
while True:
while self.active and self.on_new_message:
def _(task: asyncio.Task[str|None]):
res = task.result()
@@ -122,18 +126,19 @@ class UnixSocket:
asyncio.create_task(self.on_new_message(line)).add_done_callback(_)
class PortSocket (UnixSocket):
def __init__(self, on_new_message: Callable[[str], Coroutine[Any, Any, Any]]):
def __init__(self):
'''
on_new_message takes 1 string argument.
It's return value gets used, if not None, to write data to the socket.
Method should be async
'''
super().__init__(on_new_message)
super().__init__()
self.host = "127.0.0.1"
self.port = random.sample(range(40000, 60000), 1)[0]
async def setup_server(self):
async def setup_server(self, on_new_message: Callable[[str], Coroutine[Any, Any, Any]]):
try:
self.on_new_message = on_new_message
self.socket = await asyncio.start_server(self._listen_for_method_call, host=self.host, port=self.port, limit=BUFFER_LIMIT)
except asyncio.CancelledError:
await self.close_socket_connection()
+7 -1
View File
@@ -101,6 +101,8 @@ class PluginManager:
self.web_app.add_routes([static("/static", path.join(path.dirname(__file__), 'static'))])
async def handle_crash(self):
if not self.reinject:
return
new_time = time()
if (new_time - self.last_webhelper_exit < 60):
self.webhelper_crash_count += 1
@@ -118,9 +120,13 @@ class PluginManager:
async def shutdown(self, _: Application):
try:
logger.info(f"Shutting down...")
logger.info("Disabling reload...")
await self.plugin_loader.disable_reload()
logger.info("Killing plugins...")
await self.plugin_loader.shutdown_plugins()
await self.ws.disconnect()
logger.info("Disconnecting from WS...")
self.reinject = False
await self.ws.disconnect()
if self.js_ctx_tab:
await self.js_ctx_tab.close_websocket()
self.js_ctx_tab = None
+34 -22
View File
@@ -1,8 +1,10 @@
from asyncio import CancelledError, Task, create_task, sleep
from asyncio import CancelledError, Task, create_task, sleep, get_event_loop, wait
from json import dumps, load, loads
from logging import getLogger
from os import path
from multiprocessing import Process
from time import time
from traceback import format_exc
from .sandboxed_plugin import SandboxedPlugin
from .messages import MethodCallRequest, SocketMessageType
@@ -42,8 +44,7 @@ class PluginWrapper:
self.sandboxed_plugin = SandboxedPlugin(self.name, self.passive, self.flags, self.file, self.plugin_directory, self.plugin_path, self.version, self.author, self.api_version)
self.proc: Process | None = None
# TODO: Maybe make LocalSocket not require on_new_message to make this cleaner
self._socket = LocalSocket(self.sandboxed_plugin.on_new_message)
self._socket = LocalSocket()
self._listener_task: Task[Any]
self._method_call_requests: Dict[str, MethodCallRequest] = {}
@@ -65,7 +66,7 @@ class PluginWrapper:
return self.name
async def _response_listener(self):
while True:
while self._socket.active:
try:
line = await self._socket.read_single_line()
if line != None:
@@ -115,29 +116,40 @@ class PluginWrapper:
return self
async def stop(self, uninstall: bool = False):
self.log.info(f"Stopping plugin {self.name}")
if self.passive:
return
if hasattr(self, "_socket"):
await self._socket.write_single_line(dumps({ "stop": True, "uninstall": uninstall }, ensure_ascii=False))
await self._socket.close_socket_connection()
if self.proc:
self.proc.join()
await self.kill_if_still_running()
if hasattr(self, "_listener_task"):
self._listener_task.cancel()
try:
start_time = time()
if self.passive:
return
_, pending = await wait([
create_task(self._socket.write_single_line(dumps({ "stop": True, "uninstall": uninstall }, ensure_ascii=False)))
], timeout=1)
if hasattr(self, "_listener_task"):
self._listener_task.cancel()
await self.kill_if_still_running()
for pending_task in pending:
pending_task.cancel()
self.log.info(f"Plugin {self.name} has been stopped in {time() - start_time:.1f}s")
except Exception as e:
self.log.error(f"Error during shutdown for plugin {self.name}: {str(e)}\n{format_exc()}")
async def kill_if_still_running(self):
time = 0
start_time = time()
sigtermed = False
while self.proc and self.proc.is_alive():
await sleep(0.1)
time += 1
if time == 100:
self.log.warn(f"Plugin {self.name} still alive 10 seconds after stop request! Sending SIGTERM!")
elapsed_time = time() - start_time
if elapsed_time >= 5 and not sigtermed:
sigtermed = True
self.log.warn(f"Plugin {self.name} still alive 5 seconds after stop request! Sending SIGTERM!")
self.terminate()
elif time == 200:
self.log.warn(f"Plugin {self.name} still alive 20 seconds after stop request! Sending SIGKILL!")
elif elapsed_time >= 10:
self.log.warn(f"Plugin {self.name} still alive 10 seconds after stop request! Sending SIGKILL!")
self.terminate(True)
await sleep(0.1)
def terminate(self, kill: bool = False):
if self.proc and self.proc.is_alive():
@@ -1,6 +1,5 @@
import sys
from os import path, environ
from signal import SIG_IGN, SIGINT, SIGTERM, getsignal, signal
from importlib.util import module_from_spec, spec_from_file_location
from json import dumps, loads
from logging import getLogger
@@ -19,8 +18,6 @@ from typing import List, TypeVar, Any
DataType = TypeVar("DataType")
original_term_handler = getsignal(SIGTERM)
class SandboxedPlugin:
def __init__(self,
name: str,
@@ -48,11 +45,6 @@ class SandboxedPlugin:
self._socket = socket
try:
# Ignore signals meant for parent Process
# TODO SURELY there's a better way to do this.
signal(SIGINT, SIG_IGN)
signal(SIGTERM, SIG_IGN)
setproctitle(f"{self.name} ({self.file})")
setthreadtitle(self.name)
@@ -120,7 +112,7 @@ class SandboxedPlugin:
get_event_loop().create_task(self.Plugin._main())
else:
get_event_loop().create_task(self.Plugin._main(self.Plugin))
get_event_loop().create_task(socket.setup_server())
get_event_loop().create_task(socket.setup_server(self.on_new_message))
except:
self.log.error("Failed to start " + self.name + "!\n" + format_exc())
sys.exit(0)
@@ -167,8 +159,6 @@ class SandboxedPlugin:
data = loads(message)
if "stop" in data:
# Incase the loader needs to terminate our process soon
signal(SIGTERM, original_term_handler)
self.log.info(f"Calling Loader unload function for {self.name}.")
await self._unload()