1############################################################################### 2# Server process to keep track of unlinked resources (like shared memory 3# segments, semaphores etc.) and clean them. 4# 5# On Unix we run a server process which keeps track of unlinked 6# resources. The server ignores SIGINT and SIGTERM and reads from a 7# pipe. Every other process of the program has a copy of the writable 8# end of the pipe, so we get EOF when all other processes have exited. 9# Then the server process unlinks any remaining resource names. 10# 11# This is important because there may be system limits for such resources: for 12# instance, the system only supports a limited number of named semaphores, and 13# shared-memory segments live in the RAM. If a python process leaks such a 14# resource, this resource will not be removed till the next reboot. Without 15# this resource tracker process, "killall python" would probably leave unlinked 16# resources. 17 18import os 19import signal 20import sys 21import threading 22import warnings 23 24from . import spawn 25from . import util 26 27__all__ = ['ensure_running', 'register', 'unregister'] 28 29_HAVE_SIGMASK = hasattr(signal, 'pthread_sigmask') 30_IGNORED_SIGNALS = (signal.SIGINT, signal.SIGTERM) 31 32_CLEANUP_FUNCS = { 33 'noop': lambda: None, 34} 35 36if os.name == 'posix': 37 import _multiprocessing 38 import _posixshmem 39 40 # Use sem_unlink() to clean up named semaphores. 41 # 42 # sem_unlink() may be missing if the Python build process detected the 43 # absence of POSIX named semaphores. In that case, no named semaphores were 44 # ever opened, so no cleanup would be necessary. 45 if hasattr(_multiprocessing, 'sem_unlink'): 46 _CLEANUP_FUNCS.update({ 47 'semaphore': _multiprocessing.sem_unlink, 48 }) 49 _CLEANUP_FUNCS.update({ 50 'shared_memory': _posixshmem.shm_unlink, 51 }) 52 53 54class ResourceTracker(object): 55 56 def __init__(self): 57 self._lock = threading.Lock() 58 self._fd = None 59 self._pid = None 60 61 def _stop(self): 62 with self._lock: 63 if self._fd is None: 64 # not running 65 return 66 67 # closing the "alive" file descriptor stops main() 68 os.close(self._fd) 69 self._fd = None 70 71 os.waitpid(self._pid, 0) 72 self._pid = None 73 74 def getfd(self): 75 self.ensure_running() 76 return self._fd 77 78 def ensure_running(self): 79 '''Make sure that resource tracker process is running. 80 81 This can be run from any process. Usually a child process will use 82 the resource created by its parent.''' 83 with self._lock: 84 if self._fd is not None: 85 # resource tracker was launched before, is it still running? 86 if self._check_alive(): 87 # => still alive 88 return 89 # => dead, launch it again 90 os.close(self._fd) 91 92 # Clean-up to avoid dangling processes. 93 try: 94 # _pid can be None if this process is a child from another 95 # python process, which has started the resource_tracker. 96 if self._pid is not None: 97 os.waitpid(self._pid, 0) 98 except ChildProcessError: 99 # The resource_tracker has already been terminated. 100 pass 101 self._fd = None 102 self._pid = None 103 104 warnings.warn('resource_tracker: process died unexpectedly, ' 105 'relaunching. Some resources might leak.') 106 107 fds_to_pass = [] 108 try: 109 fds_to_pass.append(sys.stderr.fileno()) 110 except Exception: 111 pass 112 cmd = 'from multiprocessing.resource_tracker import main;main(%d)' 113 r, w = os.pipe() 114 try: 115 fds_to_pass.append(r) 116 # process will out live us, so no need to wait on pid 117 exe = spawn.get_executable() 118 args = [exe] + util._args_from_interpreter_flags() 119 args += ['-c', cmd % r] 120 # bpo-33613: Register a signal mask that will block the signals. 121 # This signal mask will be inherited by the child that is going 122 # to be spawned and will protect the child from a race condition 123 # that can make the child die before it registers signal handlers 124 # for SIGINT and SIGTERM. The mask is unregistered after spawning 125 # the child. 126 try: 127 if _HAVE_SIGMASK: 128 signal.pthread_sigmask(signal.SIG_BLOCK, _IGNORED_SIGNALS) 129 pid = util.spawnv_passfds(exe, args, fds_to_pass) 130 finally: 131 if _HAVE_SIGMASK: 132 signal.pthread_sigmask(signal.SIG_UNBLOCK, _IGNORED_SIGNALS) 133 except: 134 os.close(w) 135 raise 136 else: 137 self._fd = w 138 self._pid = pid 139 finally: 140 os.close(r) 141 142 def _check_alive(self): 143 '''Check that the pipe has not been closed by sending a probe.''' 144 try: 145 # We cannot use send here as it calls ensure_running, creating 146 # a cycle. 147 os.write(self._fd, b'PROBE:0:noop\n') 148 except OSError: 149 return False 150 else: 151 return True 152 153 def register(self, name, rtype): 154 '''Register name of resource with resource tracker.''' 155 self._send('REGISTER', name, rtype) 156 157 def unregister(self, name, rtype): 158 '''Unregister name of resource with resource tracker.''' 159 self._send('UNREGISTER', name, rtype) 160 161 def _send(self, cmd, name, rtype): 162 self.ensure_running() 163 msg = '{0}:{1}:{2}\n'.format(cmd, name, rtype).encode('ascii') 164 if len(name) > 512: 165 # posix guarantees that writes to a pipe of less than PIPE_BUF 166 # bytes are atomic, and that PIPE_BUF >= 512 167 raise ValueError('name too long') 168 nbytes = os.write(self._fd, msg) 169 assert nbytes == len(msg), "nbytes {0:n} but len(msg) {1:n}".format( 170 nbytes, len(msg)) 171 172 173_resource_tracker = ResourceTracker() 174ensure_running = _resource_tracker.ensure_running 175register = _resource_tracker.register 176unregister = _resource_tracker.unregister 177getfd = _resource_tracker.getfd 178 179def main(fd): 180 '''Run resource tracker.''' 181 # protect the process from ^C and "killall python" etc 182 signal.signal(signal.SIGINT, signal.SIG_IGN) 183 signal.signal(signal.SIGTERM, signal.SIG_IGN) 184 if _HAVE_SIGMASK: 185 signal.pthread_sigmask(signal.SIG_UNBLOCK, _IGNORED_SIGNALS) 186 187 for f in (sys.stdin, sys.stdout): 188 try: 189 f.close() 190 except Exception: 191 pass 192 193 cache = {rtype: set() for rtype in _CLEANUP_FUNCS.keys()} 194 try: 195 # keep track of registered/unregistered resources 196 with open(fd, 'rb') as f: 197 for line in f: 198 try: 199 cmd, name, rtype = line.strip().decode('ascii').split(':') 200 cleanup_func = _CLEANUP_FUNCS.get(rtype, None) 201 if cleanup_func is None: 202 raise ValueError( 203 f'Cannot register {name} for automatic cleanup: ' 204 f'unknown resource type {rtype}') 205 206 if cmd == 'REGISTER': 207 cache[rtype].add(name) 208 elif cmd == 'UNREGISTER': 209 cache[rtype].remove(name) 210 elif cmd == 'PROBE': 211 pass 212 else: 213 raise RuntimeError('unrecognized command %r' % cmd) 214 except Exception: 215 try: 216 sys.excepthook(*sys.exc_info()) 217 except: 218 pass 219 finally: 220 # all processes have terminated; cleanup any remaining resources 221 for rtype, rtype_cache in cache.items(): 222 if rtype_cache: 223 try: 224 warnings.warn('resource_tracker: There appear to be %d ' 225 'leaked %s objects to clean up at shutdown' % 226 (len(rtype_cache), rtype)) 227 except Exception: 228 pass 229 for name in rtype_cache: 230 # For some reason the process which created and registered this 231 # resource has failed to unregister it. Presumably it has 232 # died. We therefore unlink it. 233 try: 234 try: 235 _CLEANUP_FUNCS[rtype](name) 236 except Exception as e: 237 warnings.warn('resource_tracker: %r: %s' % (name, e)) 238 finally: 239 pass 240