python port of postmaster pid file loop

postmasterpid.py

#!/usr/bin/env python
import argparse
import collections
import daemon
import datetime
import errno
import os
import re
import subprocess
import sys
import time


def create_or_clean_pidfile(sentinel_path):
    my_pid = os.getpid()

    # adapted from misinit.c in postgresql
    fd = None

    # Retry lock-file cleanup/acquisition a large but finite number of
    # times.
    for i in xrange(100):
        # Try to create a lockfile.  If the lockfile is created
        # successfully, write the current PID into it.
        try:
            fd = os.open(sentinel_path,
                         os.O_CREAT | os.O_EXCL | os.O_RDWR, 0600)
            os.write(fd, str(my_pid) + '\n')
            return True
        except EnvironmentError as e:
            if e.errno in (errno.EEXIST, errno.EACCES):
                print >>sys.stderr, 'Sentinel file already exists.'
        finally:
            if fd is not None:
                os.close(fd)

        # Sentinel file was found to exist, so try reading it and
        # seeing if it is stale or not.
        try:
            with open(sentinel_path, 'r') as f:
                f.seek(0, os.SEEK_END)
                if f.tell() > 8192:
                    raise Exception('Large (>8KB) pid file found; '
                                    'that should not happen.')
                f.seek(0, os.SEEK_SET)

                pid = f.read(8192).split('\n')[0]

                try:
                    os.kill(int(pid), 0)
                    raise Exception('Is another WAL-E running? pid={0}'
                                    .format(pid))
                except EnvironmentError as e:
                    if e.errno in (errno.ESRCH, errno.EPERM):
                        # No process found, unlink the file (aka
                        # cleanup).  This is the case both if there is
                        # no pid in existence referenced by this
                        # pidfile, or if there is a permission
                        # problem, meaning an unrelated process from
                        # some other user has taken that pid (unless
                        # that other user is *also* running wal-e).
                        os.unlink(sentinel_path)
        except EnvironmentError as e:
            if e.errno == errno.ENOENT:
                # File not found; this can happen in a race condition
                # whereby another push-backup unlinks the sentinel
                # file.  Go ahead and try the entire sentinel file
                # process again.
                continue

    raise Exception('Could not create, clean, nor confirm the liveness of '
                    'another WAL-E: giving up on retry-loop.')