|
| 1 | +import asyncio |
| 2 | +import logging |
| 3 | +import os |
| 4 | +import signal |
| 5 | +import time |
| 6 | +from dataclasses import dataclass |
| 7 | +from datetime import datetime |
| 8 | +from threading import Thread |
| 9 | +from typing import Literal |
| 10 | + |
| 11 | +from aiohttp import web |
| 12 | +from aiohttp.web import Request, Response |
| 13 | +from setproctitle import setproctitle |
| 14 | + |
| 15 | + |
| 16 | +@dataclass |
| 17 | +class KeepAliveConfig: |
| 18 | + listen_address: str = "127.0.0.1" |
| 19 | + listen_port: int = 18080 |
| 20 | + |
| 21 | + watchdog_feed_interval: int = 600 |
| 22 | + |
| 23 | + keep_alive_method: Literal["special_process"] = "special_process" |
| 24 | + |
| 25 | + special_process_name: str | None = "wsp-keepalive" |
| 26 | + |
| 27 | + |
| 28 | +class KeepAliveDaemon: |
| 29 | + _log: logging.Logger = logging.getLogger(__name__) |
| 30 | + |
| 31 | + _config: KeepAliveConfig |
| 32 | + _web_app: web.Application |
| 33 | + |
| 34 | + _watchdog_last_feed_time: datetime |
| 35 | + _watchdog_is_hungry: bool = False |
| 36 | + |
| 37 | + _special_process_id: int = -1 |
| 38 | + |
| 39 | + def __init__(self, config: KeepAliveConfig) -> None: |
| 40 | + self._config = config |
| 41 | + |
| 42 | + self._web_app = web.Application() |
| 43 | + |
| 44 | + self._web_app.add_routes( |
| 45 | + [ |
| 46 | + web.get("/watchdog/feed", self._handle_watchdog_feed) |
| 47 | + ] |
| 48 | + ) |
| 49 | + |
| 50 | + async def _watchdog_timer(self) -> None: |
| 51 | + while True: |
| 52 | + previous_is_hungry = self._watchdog_is_hungry |
| 53 | + last_feed_interval = datetime.now() - self._watchdog_last_feed_time |
| 54 | + |
| 55 | + if last_feed_interval.total_seconds() > self._config.watchdog_feed_interval: |
| 56 | + self._watchdog_is_hungry = True |
| 57 | + else: |
| 58 | + self._watchdog_is_hungry = False |
| 59 | + |
| 60 | + if not previous_is_hungry and self._watchdog_is_hungry: |
| 61 | + Thread(target=self._on_watchdog_hungry, daemon=True).start() |
| 62 | + elif previous_is_hungry and not self._watchdog_is_hungry: |
| 63 | + Thread(target=self._on_watchdog_feed, daemon=True).start() |
| 64 | + |
| 65 | + await asyncio.sleep(1) |
| 66 | + |
| 67 | + def _on_watchdog_hungry(self) -> None: |
| 68 | + self._log.warning("Watchdog is hungry!") |
| 69 | + |
| 70 | + if self._special_process_id > 0: |
| 71 | + os.kill(self._special_process_id, signal.SIGKILL) |
| 72 | + os.waitpid(self._special_process_id, 0) |
| 73 | + self._log.warning("Killed special process PID %d", self._special_process_id) |
| 74 | + self._special_process_id = -1 |
| 75 | + |
| 76 | + def _on_watchdog_feed(self) -> None: |
| 77 | + self._log.info("Watchdog is feed.") |
| 78 | + |
| 79 | + if self._config.keep_alive_method == "special_process" and self._special_process_id < 0: |
| 80 | + self._start_special_process() |
| 81 | + |
| 82 | + def _special_process(self) -> None: |
| 83 | + assert self._config.special_process_name is not None |
| 84 | + setproctitle(self._config.special_process_name) |
| 85 | + |
| 86 | + while True: |
| 87 | + time.sleep(60) |
| 88 | + |
| 89 | + if os.getppid() == 1: |
| 90 | + os._exit(0) |
| 91 | + |
| 92 | + def _start_special_process(self) -> bool: |
| 93 | + self._special_process_id = os.fork() |
| 94 | + |
| 95 | + if self._special_process_id == 0: |
| 96 | + self._special_process() |
| 97 | + return True |
| 98 | + |
| 99 | + self._log.info( |
| 100 | + "Started special process with name %s, PID %d", |
| 101 | + self._config.special_process_name, self._special_process_id |
| 102 | + ) |
| 103 | + |
| 104 | + return False |
| 105 | + |
| 106 | + def start(self) -> None: |
| 107 | + self._watchdog_last_feed_time = datetime.now() |
| 108 | + self._watchdog_is_hungry = False |
| 109 | + |
| 110 | + if self._config.keep_alive_method == "special_process": # noqa: SIM102 |
| 111 | + if self._start_special_process(): |
| 112 | + return |
| 113 | + |
| 114 | + loop = asyncio.new_event_loop() |
| 115 | + |
| 116 | + def _start_loop() -> None: |
| 117 | + asyncio.set_event_loop(loop) |
| 118 | + loop.run_forever() |
| 119 | + |
| 120 | + Thread(target=_start_loop, daemon=True).start() |
| 121 | + |
| 122 | + watchdog_task = asyncio.run_coroutine_threadsafe(self._watchdog_timer(), loop) |
| 123 | + |
| 124 | + self._log.info( |
| 125 | + "Keep-alive daemon started at %s:%d, watchdog feed interval %ds.", |
| 126 | + self._config.listen_address, self._config.listen_port, self._config.watchdog_feed_interval |
| 127 | + ) |
| 128 | + |
| 129 | + web.run_app(self._web_app, host=self._config.listen_address, port=self._config.listen_port) |
| 130 | + |
| 131 | + self._log.info("Stopping...") |
| 132 | + |
| 133 | + watchdog_task.cancel() |
| 134 | + loop.stop() |
| 135 | + |
| 136 | + if self._special_process_id > 0: |
| 137 | + os.kill(self._special_process_id, signal.SIGKILL) |
| 138 | + os.waitpid(self._special_process_id, 0) |
| 139 | + |
| 140 | + self._log.info("Stopped.") |
| 141 | + |
| 142 | + async def _handle_watchdog_feed(self, _: Request) -> Response: |
| 143 | + last_feed = self._watchdog_last_feed_time |
| 144 | + self._watchdog_last_feed_time = datetime.now() |
| 145 | + interval = (self._watchdog_last_feed_time - last_feed).total_seconds() |
| 146 | + |
| 147 | + return web.json_response( |
| 148 | + { |
| 149 | + "status": "ok", |
| 150 | + "last_feed": last_feed.strftime("%Y-%m-%d %H:%M:%S"), |
| 151 | + "interval": interval, |
| 152 | + } |
| 153 | + ) |
0 commit comments