relay_client_udp.py 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252
  1. from __future__ import annotations
  2. import asyncio
  3. import contextlib
  4. import json
  5. import random
  6. import socket
  7. from dataclasses import dataclass
  8. import time
  9. from typing import Awaitable, Callable, Dict
  10. from .config_udp import UdpConfig, UdpRelayNode
  11. from .protocol import AUTH, PING, PONG, STATUS_OK, TCP_CLOSE, Frame, encode_json, read_frame, write_frame
  12. from .scheduler_udp import UdpScheduler
  13. FrameHandler = Callable[["UdpRelayConnection", Frame], Awaitable[None]]
  14. @dataclass
  15. class UdpRelayConnection:
  16. node: UdpRelayNode
  17. manager: "UdpRelayManager"
  18. reader: asyncio.StreamReader
  19. writer: asyncio.StreamWriter
  20. closed: bool = False
  21. handlers: Dict[tuple[int, int], FrameHandler] = None
  22. dispatch_tasks: Dict[tuple[int, int], asyncio.Task] = None
  23. pump_task: asyncio.Task | None = None
  24. keepalive_task: asyncio.Task | None = None
  25. last_pong_at: float = 0.0
  26. send_lock: asyncio.Lock | None = None
  27. closed_event: asyncio.Event | None = None
  28. dropped_frames: Dict[int, int] = None
  29. dropped_report_task: asyncio.Task | None = None
  30. def __post_init__(self) -> None:
  31. if self.handlers is None:
  32. self.handlers = {}
  33. if self.dispatch_tasks is None:
  34. self.dispatch_tasks = {}
  35. if self.send_lock is None:
  36. self.send_lock = asyncio.Lock()
  37. if self.closed_event is None:
  38. self.closed_event = asyncio.Event()
  39. if self.dropped_frames is None:
  40. self.dropped_frames = {}
  41. async def start(self) -> None:
  42. print(f"[edge] connecting relay name={self.node.name} addr={self.node.host}:{self.node.port}")
  43. await write_frame(self.writer, Frame(AUTH, 0, 0, 0, 0, encode_json({"token": self.node.token})))
  44. frame = await read_frame(self.reader)
  45. if frame.kind != AUTH or frame.packet_id != STATUS_OK:
  46. raise ConnectionError(f"relay auth failed: {self.node.name}")
  47. if frame.payload:
  48. with contextlib.suppress(Exception):
  49. json.loads(frame.payload.decode("utf-8"))
  50. self.last_pong_at = time.monotonic()
  51. self.keepalive_task = asyncio.create_task(self._keepalive())
  52. self.pump_task = asyncio.create_task(self._pump())
  53. print(f"[edge] relay connected name={self.node.name} addr={self.node.host}:{self.node.port} mode=udp")
  54. async def _keepalive(self) -> None:
  55. try:
  56. while not self.closed:
  57. await asyncio.sleep(self.manager.config.relay_ping_interval)
  58. if self.closed:
  59. break
  60. if self.last_pong_at and time.monotonic() - self.last_pong_at > (self.manager.config.relay_ping_interval + self.manager.config.relay_ping_timeout):
  61. print(f"[edge] relay health timeout name={self.node.name} addr={self.node.host}:{self.node.port} timeout={self.manager.config.relay_ping_timeout}")
  62. await self.close()
  63. break
  64. await self.send(Frame(PING, 0, 0, 0, 0, b""))
  65. except asyncio.CancelledError:
  66. pass
  67. except Exception:
  68. await self.close()
  69. async def _pump(self) -> None:
  70. try:
  71. while True:
  72. frame = await read_frame(self.reader)
  73. if frame.kind == PONG:
  74. self.last_pong_at = time.monotonic()
  75. continue
  76. handler = self.handlers.get((frame.session_id, frame.stream_id))
  77. if handler:
  78. self._dispatch_frame(frame, handler)
  79. else:
  80. self._record_dropped_frame(frame.kind)
  81. except (asyncio.IncompleteReadError, ConnectionResetError, BrokenPipeError, OSError):
  82. pass
  83. except Exception:
  84. pass
  85. finally:
  86. await self.close()
  87. def _dispatch_frame(self, frame: Frame, handler: FrameHandler) -> None:
  88. key = (frame.session_id, frame.stream_id)
  89. previous = self.dispatch_tasks.get(key)
  90. task = asyncio.create_task(self._run_handler(key, frame, handler, previous))
  91. self.dispatch_tasks[key] = task
  92. async def _run_handler(self, key: tuple[int, int], frame: Frame, handler: FrameHandler, previous: asyncio.Task | None) -> None:
  93. try:
  94. if previous is not None:
  95. with contextlib.suppress(Exception):
  96. await previous
  97. if self.closed:
  98. return
  99. await handler(self, frame)
  100. except asyncio.CancelledError:
  101. pass
  102. except Exception:
  103. if not self.closed:
  104. await self.close()
  105. finally:
  106. if self.dispatch_tasks.get(key) is asyncio.current_task():
  107. self.dispatch_tasks.pop(key, None)
  108. def _record_dropped_frame(self, kind: int) -> None:
  109. self.dropped_frames[kind] = self.dropped_frames.get(kind, 0) + 1
  110. if self.dropped_report_task is None or self.dropped_report_task.done():
  111. self.dropped_report_task = asyncio.create_task(self._report_dropped_frames())
  112. async def _report_dropped_frames(self) -> None:
  113. try:
  114. await asyncio.sleep(5)
  115. dropped = self.dropped_frames
  116. self.dropped_frames = {}
  117. if dropped:
  118. detail = ", ".join(f"kind={kind} count={count}" for kind, count in sorted(dropped.items()))
  119. print(f"[edge] relay frame dropped summary name={self.node.name} {detail}")
  120. except asyncio.CancelledError:
  121. pass
  122. async def send(self, frame: Frame) -> None:
  123. if self.closed:
  124. raise ConnectionError(f"relay closed: {self.node.name}")
  125. assert self.send_lock is not None
  126. async with self.send_lock:
  127. if self.closed:
  128. raise ConnectionError(f"relay closed: {self.node.name}")
  129. await write_frame(self.writer, frame)
  130. def bind(self, session_id: int, stream_id: int, handler: FrameHandler) -> None:
  131. self.handlers[(session_id, stream_id)] = handler
  132. def unbind(self, session_id: int, stream_id: int) -> None:
  133. self.handlers.pop((session_id, stream_id), None)
  134. task = self.dispatch_tasks.pop((session_id, stream_id), None)
  135. if task is not None:
  136. task.cancel()
  137. async def close(self) -> None:
  138. if self.closed:
  139. return
  140. self.closed = True
  141. assert self.closed_event is not None
  142. self.closed_event.set()
  143. handlers = list(self.handlers.items())
  144. self.handlers.clear()
  145. dispatch_tasks = list(self.dispatch_tasks.values())
  146. self.dispatch_tasks.clear()
  147. self.manager.on_closed(self)
  148. for (session_id, stream_id), handler in handlers:
  149. with contextlib.suppress(Exception):
  150. await handler(self, Frame(TCP_CLOSE, session_id, stream_id, 0, 0, b""))
  151. for task in dispatch_tasks:
  152. task.cancel()
  153. for task in dispatch_tasks:
  154. with contextlib.suppress(Exception):
  155. await task
  156. if self.dropped_report_task and self.dropped_report_task is not asyncio.current_task():
  157. self.dropped_report_task.cancel()
  158. with contextlib.suppress(Exception):
  159. await self.dropped_report_task
  160. if self.pump_task and self.pump_task is not asyncio.current_task():
  161. self.pump_task.cancel()
  162. with contextlib.suppress(Exception):
  163. await self.pump_task
  164. if self.keepalive_task and self.keepalive_task is not asyncio.current_task():
  165. self.keepalive_task.cancel()
  166. with contextlib.suppress(Exception):
  167. await self.keepalive_task
  168. self.writer.close()
  169. with contextlib.suppress(Exception):
  170. await self.writer.wait_closed()
  171. class UdpRelayManager:
  172. def __init__(self, config: UdpConfig) -> None:
  173. self.config = config
  174. self.scheduler = UdpScheduler(config)
  175. self.connections: Dict[str, UdpRelayConnection] = {}
  176. self.tasks: list[asyncio.Task] = []
  177. async def start(self) -> None:
  178. await self.scheduler.start()
  179. for node in self.config.relays:
  180. self.tasks.append(asyncio.create_task(self._maintain(node)))
  181. async def _maintain(self, node: UdpRelayNode) -> None:
  182. backoff = self.config.relay_reconnect_delay
  183. while True:
  184. current = self.connections.get(node.name)
  185. if current is not None and not current.closed:
  186. assert current.closed_event is not None
  187. await current.closed_event.wait()
  188. continue
  189. attempt = 1
  190. while True:
  191. try:
  192. print(f"[edge] relay reconnect attempt name={node.name} addr={node.host}:{node.port} attempt={attempt} backoff={backoff:.1f}s")
  193. reader, writer = await asyncio.wait_for(asyncio.open_connection(node.host, node.port), timeout=self.config.relay_open_timeout)
  194. connection = UdpRelayConnection(node=node, manager=self, reader=reader, writer=writer)
  195. sock = writer.get_extra_info("socket")
  196. if sock is not None and self.config.relay_tcp_nodelay:
  197. with contextlib.suppress(OSError):
  198. sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)
  199. await connection.start()
  200. self.connections[node.name] = connection
  201. backoff = self.config.relay_reconnect_delay
  202. assert connection.closed_event is not None
  203. await connection.closed_event.wait()
  204. print(f"[edge] relay supervisor noticed close name={node.name} addr={node.host}:{node.port}")
  205. break
  206. except asyncio.CancelledError:
  207. raise
  208. except Exception as exc:
  209. print(f"[edge] relay connect failed name={node.name} addr={node.host}:{node.port} attempt={attempt} error={exc!r}")
  210. jitter = random.uniform(0, min(1.0, backoff * 0.2))
  211. await asyncio.sleep(backoff + jitter)
  212. backoff = min(self.config.relay_reconnect_max_delay, max(self.config.relay_reconnect_delay, backoff * 2))
  213. attempt += 1
  214. def on_closed(self, connection: UdpRelayConnection) -> None:
  215. current = self.connections.get(connection.node.name)
  216. if current is connection:
  217. self.connections.pop(connection.node.name, None)
  218. def available_udp(self) -> list[UdpRelayConnection]:
  219. chosen = {node.name for node in self.scheduler.choose()}
  220. preferred = [self.connections[name] for name in chosen if name in self.connections and not self.connections[name].closed]
  221. if preferred:
  222. return preferred
  223. return [conn for conn in self.connections.values() if not conn.closed]
  224. def snapshot(self) -> list[dict[str, object]]:
  225. data = self.scheduler.snapshot()
  226. online = {name for name, conn in self.connections.items() if not conn.closed}
  227. for item in data:
  228. item["online"] = item["name"] in online
  229. return data