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