lockfile.py

   1 #!/usr/bin/env python3
   2
   3 from dataclasses import dataclass
   4 import datetime
   5 import json
   6 import logging
   7 import os
   8 import signal
   9 import sys
  10 from typing import Optional
  11
  12 import config
  13 import datetime_utils
  14 import decorator_utils
  15
  16
  17 cfg = config.add_commandline_args(
  18     f'Lockfile ({__file__})',
  19     'Args related to lockfiles')
  20 cfg.add_argument(
  21     '--lockfile_held_duration_warning_threshold_sec',
  22     type=float,
  23     default=10.0,
  24     metavar='SECONDS',
  25     help='If a lock is held for longer than this threshold we log a warning'
  26 )
  27 logger = logging.getLogger(__name__)
  28
  29
  30 class LockFileException(Exception):
  31     pass
  32
  33
  34 @dataclass
  35 class LockFileContents:
  36     pid: int
  37     commandline: str
  38     expiration_timestamp: float
  39
  40
  41 class LockFile(object):
  42     """A file locking mechanism that has context-manager support so you
  43     can use it in a with statement.  e.g.
  44
  45     with LockFile('./foo.lock'):
  46         # do a bunch of stuff... if the process dies we have a signal
  47         # handler to do cleanup.  Other code (in this process or another)
  48         # that tries to take the same lockfile will block.  There is also
  49         # some logic for detecting stale locks.
  50
  51     """
  52     def __init__(
  53             self,
  54             lockfile_path: str,
  55             *,
  56             do_signal_cleanup: bool = True,
  57             expiration_timestamp: Optional[float] = None,
  58             override_command: Optional[str] = None,
  59     ) -> None:
  60         self.is_locked = False
  61         self.lockfile = lockfile_path
  62         self.override_command = override_command
  63         if do_signal_cleanup:
  64             signal.signal(signal.SIGINT, self._signal)
  65             signal.signal(signal.SIGTERM, self._signal)
  66         self.expiration_timestamp = expiration_timestamp
  67
  68     def locked(self):
  69         return self.is_locked
  70
  71     def available(self):
  72         return not os.path.exists(self.lockfile)
  73
  74     def try_acquire_lock_once(self) -> bool:
  75         logger.debug(f"Trying to acquire {self.lockfile}.")
  76         try:
  77             # Attempt to create the lockfile.  These flags cause
  78             # os.open to raise an OSError if the file already
  79             # exists.
  80             fd = os.open(self.lockfile, os.O_CREAT | os.O_EXCL | os.O_RDWR)
  81             with os.fdopen(fd, "a") as f:
  82                 contents = self._get_lockfile_contents()
  83                 logger.debug(contents)
  84                 f.write(contents)
  85             logger.debug(f'Success; I own {self.lockfile}.')
  86             self.is_locked = True
  87             return True
  88         except OSError:
  89             pass
  90         logger.warning(f'Could not acquire {self.lockfile}.')
  91         return False
  92
  93     def acquire_with_retries(
  94             self,
  95             *,
  96             initial_delay: float = 1.0,
  97             backoff_factor: float = 2.0,
  98             max_attempts = 5
  99     ) -> bool:
 100
 101         @decorator_utils.retry_if_false(tries = max_attempts,
 102                                         delay_sec = initial_delay,
 103                                         backoff = backoff_factor)
 104         def _try_acquire_lock_with_retries() -> bool:
 105             success = self.try_acquire_lock_once()
 106             if not success and os.path.exists(self.lockfile):
 107                 self._detect_stale_lockfile()
 108             return success
 109
 110         if os.path.exists(self.lockfile):
 111             self._detect_stale_lockfile()
 112         return _try_acquire_lock_with_retries()
 113
 114     def release(self):
 115         try:
 116             os.unlink(self.lockfile)
 117         except Exception as e:
 118             logger.exception(e)
 119         self.is_locked = False
 120
 121     def __enter__(self):
 122         if self.acquire_with_retries():
 123             self.locktime = datetime.datetime.now().timestamp()
 124             return self
 125         msg = f"Couldn't acquire {self.lockfile}; giving up."
 126         logger.warning(msg)
 127         raise LockFileException(msg)
 128
 129     def __exit__(self, type, value, traceback):
 130         if self.locktime:
 131             ts = datetime.datetime.now().timestamp()
 132             duration = ts - self.locktime
 133             if duration >= config.config['lockfile_held_duration_warning_threshold_sec']:
 134                 str_duration = datetime_utils.describe_duration_briefly(duration)
 135                 logger.warning(f'Held {self.lockfile} for {str_duration}')
 136         self.release()
 137
 138     def __del__(self):
 139         if self.is_locked:
 140             self.release()
 141
 142     def _signal(self, *args):
 143         if self.is_locked:
 144             self.release()
 145
 146     def _get_lockfile_contents(self) -> str:
 147         if self.override_command:
 148             cmd = self.override_command
 149         else:
 150             cmd = ' '.join(sys.argv)
 151         print(cmd)
 152         contents = LockFileContents(
 153             pid = os.getpid(),
 154             commandline = cmd,
 155             expiration_timestamp = self.expiration_timestamp,
 156         )
 157         return json.dumps(contents.__dict__)
 158
 159     def _detect_stale_lockfile(self) -> None:
 160         try:
 161             with open(self.lockfile, 'r') as rf:
 162                 lines = rf.readlines()
 163                 if len(lines) == 1:
 164                     line = lines[0]
 165                     line_dict = json.loads(line)
 166                     contents = LockFileContents(**line_dict)
 167                     logger.debug(f'Blocking lock contents="{contents}"')
 168
 169                     # Does the PID exist still?
 170                     try:
 171                         os.kill(contents.pid, 0)
 172                     except OSError:
 173                         logger.warning(f'Lockfile {self.lockfile}\'s pid ({contents.pid}) is stale; ' +
 174                                        'force acquiring')
 175                         self.release()
 176
 177                     # Has the lock expiration expired?
 178                     if contents.expiration_timestamp is not None:
 179                         now = datetime.datetime.now().timestamp()
 180                         if now > contents.expiration_datetime:
 181                             logger.warning(f'Lockfile {self.lockfile} expiration time has passed; ' +
 182                                            'force acquiring')
 183                             self.release()
 184         except Exception:
 185             pass