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