path: root/cherrypy/process/wspbus.py

r"""An implementation of the Web Site Process Bus.

This module is completely standalone, depending only on the stdlib.

Web Site Process Bus
--------------------

A Bus object is used to contain and manage site-wide behavior:
daemonization, HTTP server start/stop, process reload, signal handling,
drop privileges, PID file management, logging for all of these,
and many more.

In addition, a Bus object provides a place for each web framework
to register code that runs in response to site-wide events (like
process start and stop), or which controls or otherwise interacts with
the site-wide components mentioned above. For example, a framework which
uses file-based templates would add known template filenames to an
autoreload component.

Ideally, a Bus object will be flexible enough to be useful in a variety
of invocation scenarios:

 1. The deployer starts a site from the command line via a
    framework-neutral deployment script; applications from multiple frameworks
    are mixed in a single site. Command-line arguments and configuration
    files are used to define site-wide components such as the HTTP server,
    WSGI component graph, autoreload behavior, signal handling, etc.
 2. The deployer starts a site via some other process, such as Apache;
    applications from multiple frameworks are mixed in a single site.
    Autoreload and signal handling (from Python at least) are disabled.
 3. The deployer starts a site via a framework-specific mechanism;
    for example, when running tests, exploring tutorials, or deploying
    single applications from a single framework. The framework controls
    which site-wide components are enabled as it sees fit.

The Bus object in this package uses topic-based publish-subscribe
messaging to accomplish all this. A few topic channels are built in
('start', 'stop', 'exit', 'graceful', 'log', and 'main'). Frameworks and
site containers are free to define their own. If a message is sent to a
channel that has not been defined or has no listeners, there is no effect.

In general, there should only ever be a single Bus object per process.
Frameworks and site containers share a single Bus object by publishing
messages and subscribing listeners.

The Bus object works as a finite state machine which models the current
state of the process. Bus methods move it from one state to another;
those methods then publish to subscribed listeners on the channel for
the new state.::

                        O
                        |
                        V
       STOPPING --> STOPPED --> EXITING -> X
          A   A         |
          |    \___     |
          |        \    |
          |         V   V
        STARTED <-- STARTING

"""

import atexit

try:
    import ctypes
except ImportError:
    """Google AppEngine is shipped without ctypes

    :seealso: http://stackoverflow.com/a/6523777/70170
    """
    ctypes = None

import operator
import os
import sys
import threading
import time
import traceback as _traceback
import warnings
import subprocess
import functools

from more_itertools import always_iterable


# Here I save the value of os.getcwd(), which, if I am imported early enough,
# will be the directory from which the startup script was run.  This is needed
# by _do_execv(), to change back to the original directory before execv()ing a
# new process.  This is a defense against the application having changed the
# current working directory (which could make sys.executable "not found" if
# sys.executable is a relative-path, and/or cause other problems).
_startup_cwd = os.getcwd()


class ChannelFailures(Exception):
    """Exception raised during errors on Bus.publish()."""

    delimiter = '\n'

    def __init__(self, *args, **kwargs):
        """Initialize ChannelFailures errors wrapper."""
        super(ChannelFailures, self).__init__(*args, **kwargs)
        self._exceptions = list()

    def handle_exception(self):
        """Append the current exception to self."""
        self._exceptions.append(sys.exc_info()[1])

    def get_instances(self):
        """Return a list of seen exception instances."""
        return self._exceptions[:]

    def __str__(self):
        """Render the list of errors, which happened in channel."""
        exception_strings = map(repr, self.get_instances())
        return self.delimiter.join(exception_strings)

    __repr__ = __str__

    def __bool__(self):
        """Determine whether any error happened in channel."""
        return bool(self._exceptions)
    __nonzero__ = __bool__

# Use a flag to indicate the state of the bus.


class _StateEnum(object):

    class State(object):
        name = None

        def __repr__(self):
            return 'states.%s' % self.name

    def __setattr__(self, key, value):
        if isinstance(value, self.State):
            value.name = key
        object.__setattr__(self, key, value)


states = _StateEnum()
states.STOPPED = states.State()
states.STARTING = states.State()
states.STARTED = states.State()
states.STOPPING = states.State()
states.EXITING = states.State()


try:
    import fcntl
except ImportError:
    max_files = 0
else:
    try:
        max_files = os.sysconf('SC_OPEN_MAX')
    except AttributeError:
        max_files = 1024


class Bus(object):
    """Process state-machine and messenger for HTTP site deployment.

    All listeners for a given channel are guaranteed to be called even
    if others at the same channel fail. Each failure is logged, but
    execution proceeds on to the next listener. The only way to stop all
    processing from inside a listener is to raise SystemExit and stop the
    whole server.
    """

    states = states
    state = states.STOPPED
    execv = False
    max_cloexec_files = max_files

    def __init__(self):
        """Initialize pub/sub bus."""
        self.execv = False
        self.state = states.STOPPED
        channels = 'start', 'stop', 'exit', 'graceful', 'log', 'main'
        self.listeners = dict(
            (channel, set())
            for channel in channels
        )
        self._priorities = {}

    def subscribe(self, channel, callback=None, priority=None):
        """Add the given callback at the given channel (if not present).

        If callback is None, return a partial suitable for decorating
        the callback.
        """
        if callback is None:
            return functools.partial(
                self.subscribe,
                channel,
                priority=priority,
            )

        ch_listeners = self.listeners.setdefault(channel, set())
        ch_listeners.add(callback)

        if priority is None:
            priority = getattr(callback, 'priority', 50)
        self._priorities[(channel, callback)] = priority

    def unsubscribe(self, channel, callback):
        """Discard the given callback (if present)."""
        listeners = self.listeners.get(channel)
        if listeners and callback in listeners:
            listeners.discard(callback)
            del self._priorities[(channel, callback)]

    def publish(self, channel, *args, **kwargs):
        """Return output of all subscribers for the given channel."""
        if channel not in self.listeners:
            return []

        exc = ChannelFailures()
        output = []

        raw_items = (
            (self._priorities[(channel, listener)], listener)
            for listener in self.listeners[channel]
        )
        items = sorted(raw_items, key=operator.itemgetter(0))
        for priority, listener in items:
            try:
                output.append(listener(*args, **kwargs))
            except KeyboardInterrupt:
                raise
            except SystemExit:
                e = sys.exc_info()[1]
                # If we have previous errors ensure the exit code is non-zero
                if exc and e.code == 0:
                    e.code = 1
                raise
            except Exception:
                exc.handle_exception()
                if channel == 'log':
                    # Assume any further messages to 'log' will fail.
                    pass
                else:
                    self.log('Error in %r listener %r' % (channel, listener),
                             level=40, traceback=True)
        if exc:
            raise exc
        return output

    def _clean_exit(self):
        """Assert that the Bus is not running in atexit handler callback."""
        if self.state != states.EXITING:
            warnings.warn(
                'The main thread is exiting, but the Bus is in the %r state; '
                'shutting it down automatically now. You must either call '
                'bus.block() after start(), or call bus.exit() before the '
                'main thread exits.' % self.state, RuntimeWarning)
            self.exit()

    def start(self):
        """Start all services."""
        atexit.register(self._clean_exit)

        self.state = states.STARTING
        self.log('Bus STARTING')
        try:
            self.publish('start')
            self.state = states.STARTED
            self.log('Bus STARTED')
        except (KeyboardInterrupt, SystemExit):
            raise
        except Exception:
            self.log('Shutting down due to error in start listener:',
                     level=40, traceback=True)
            e_info = sys.exc_info()[1]
            try:
                self.exit()
            except Exception:
                # Any stop/exit errors will be logged inside publish().
                pass
            # Re-raise the original error
            raise e_info

    def exit(self):
        """Stop all services and prepare to exit the process."""
        exitstate = self.state
        EX_SOFTWARE = 70
        try:
            self.stop()

            self.state = states.EXITING
            self.log('Bus EXITING')
            self.publish('exit')
            # This isn't strictly necessary, but it's better than seeing
            # "Waiting for child threads to terminate..." and then nothing.
            self.log('Bus EXITED')
        except Exception:
            # This method is often called asynchronously (whether thread,
            # signal handler, console handler, or atexit handler), so we
            # can't just let exceptions propagate out unhandled.
            # Assume it's been logged and just die.
            os._exit(EX_SOFTWARE)

        if exitstate == states.STARTING:
            # exit() was called before start() finished, possibly due to
            # Ctrl-C because a start listener got stuck. In this case,
            # we could get stuck in a loop where Ctrl-C never exits the
            # process, so we just call os.exit here.
            os._exit(EX_SOFTWARE)

    def restart(self):
        """Restart the process (may close connections).

        This method does not restart the process from the calling thread;
        instead, it stops the bus and asks the main thread to call execv.
        """
        self.execv = True
        self.exit()

    def graceful(self):
        """Advise all services to reload."""
        self.log('Bus graceful')
        self.publish('graceful')

    def block(self, interval=0.1):
        """Wait for the EXITING state, KeyboardInterrupt or SystemExit.

        This function is intended to be called only by the main thread.
        After waiting for the EXITING state, it also waits for all threads
        to terminate, and then calls os.execv if self.execv is True. This
        design allows another thread to call bus.restart, yet have the main
        thread perform the actual execv call (required on some platforms).
        """
        try:
            self.wait(states.EXITING, interval=interval, channel='main')
        except (KeyboardInterrupt, IOError):
            # The time.sleep call might raise
            # "IOError: [Errno 4] Interrupted function call" on KBInt.
            self.log('Keyboard Interrupt: shutting down bus')
            self.exit()
        except SystemExit:
            self.log('SystemExit raised: shutting down bus')
            self.exit()
            raise

        # Waiting for ALL child threads to finish is necessary on OS X.
        # See https://github.com/cherrypy/cherrypy/issues/581.
        # It's also good to let them all shut down before allowing
        # the main thread to call atexit handlers.
        # See https://github.com/cherrypy/cherrypy/issues/751.
        self.log('Waiting for child threads to terminate...')
        for t in threading.enumerate():
            # Validate the we're not trying to join the MainThread
            # that will cause a deadlock and the case exist when
            # implemented as a windows service and in any other case
            # that another thread executes cherrypy.engine.exit()
            if (
                    t != threading.currentThread() and
                    not isinstance(t, threading._MainThread) and
                    # Note that any dummy (external) threads are
                    # always daemonic.
                    not t.daemon
            ):
                self.log('Waiting for thread %s.' % t.getName())
                t.join()

        if self.execv:
            self._do_execv()

    def wait(self, state, interval=0.1, channel=None):
        """Poll for the given state(s) at intervals; publish to channel."""
        states = set(always_iterable(state))

        while self.state not in states:
            time.sleep(interval)
            self.publish(channel)

    def _do_execv(self):
        """Re-execute the current process.

        This must be called from the main thread, because certain platforms
        (OS X) don't allow execv to be called in a child thread very well.
        """
        try:
            args = self._get_true_argv()
        except NotImplementedError:
            """It's probably win32 or GAE"""
            args = [sys.executable] + self._get_interpreter_argv() + sys.argv

        self.log('Re-spawning %s' % ' '.join(args))

        self._extend_pythonpath(os.environ)

        if sys.platform[:4] == 'java':
            from _systemrestart import SystemRestart
            raise SystemRestart
        else:
            if sys.platform == 'win32':
                args = ['"%s"' % arg for arg in args]

            os.chdir(_startup_cwd)
            if self.max_cloexec_files:
                self._set_cloexec()
            os.execv(sys.executable, args)

    @staticmethod
    def _get_interpreter_argv():
        """Retrieve current Python interpreter's arguments.

        Returns empty tuple in case of frozen mode, uses built-in arguments
        reproduction function otherwise.

        Frozen mode is possible for the app has been packaged into a binary
        executable using py2exe. In this case the interpreter's arguments are
        already built-in into that executable.

        :seealso: https://github.com/cherrypy/cherrypy/issues/1526
        Ref: https://pythonhosted.org/PyInstaller/runtime-information.html
        """
        return ([]
                if getattr(sys, 'frozen', False)
                else subprocess._args_from_interpreter_flags())

    @staticmethod
    def _get_true_argv():
        """Retrieve all real arguments of the python interpreter.

        ...even those not listed in ``sys.argv``

        :seealso: http://stackoverflow.com/a/28338254/595220
        :seealso: http://stackoverflow.com/a/6683222/595220
        :seealso: http://stackoverflow.com/a/28414807/595220
        """
        try:
            char_p = ctypes.c_wchar_p

            argv = ctypes.POINTER(char_p)()
            argc = ctypes.c_int()

            ctypes.pythonapi.Py_GetArgcArgv(
                ctypes.byref(argc),
                ctypes.byref(argv),
            )

            _argv = argv[:argc.value]

            # The code below is trying to correctly handle special cases.
            # `-c`'s argument interpreted by Python itself becomes `-c` as
            # well. Same applies to `-m`. This snippet is trying to survive
            # at least the case with `-m`
            # Ref: https://github.com/cherrypy/cherrypy/issues/1545
            # Ref: python/cpython@418baf9
            argv_len, is_command, is_module = len(_argv), False, False

            try:
                m_ind = _argv.index('-m')
                if m_ind < argv_len - 1 and _argv[m_ind + 1] in ('-c', '-m'):
                    """
                    In some older Python versions `-m`'s argument may be
                    substituted with `-c`, not `-m`
                    """
                    is_module = True
            except (IndexError, ValueError):
                m_ind = None

            try:
                c_ind = _argv.index('-c')
                if c_ind < argv_len - 1 and _argv[c_ind + 1] == '-c':
                    is_command = True
            except (IndexError, ValueError):
                c_ind = None

            if is_module:
                """It's containing `-m -m` sequence of arguments"""
                if is_command and c_ind < m_ind:
                    """There's `-c -c` before `-m`"""
                    raise RuntimeError(
                        "Cannot reconstruct command from '-c'. Ref: "
                        'https://github.com/cherrypy/cherrypy/issues/1545')
                # Survive module argument here
                original_module = sys.argv[0]
                if not os.access(original_module, os.R_OK):
                    """There's no such module exist"""
                    raise AttributeError(
                        "{} doesn't seem to be a module "
                        'accessible by current user'.format(original_module))
                del _argv[m_ind:m_ind + 2]  # remove `-m -m`
                # ... and substitute it with the original module path:
                _argv.insert(m_ind, original_module)
            elif is_command:
                """It's containing just `-c -c` sequence of arguments"""
                raise RuntimeError(
                    "Cannot reconstruct command from '-c'. "
                    'Ref: https://github.com/cherrypy/cherrypy/issues/1545')
        except AttributeError:
            """It looks Py_GetArgcArgv is completely absent in some environments

            It is known, that there's no Py_GetArgcArgv in MS Windows and
            ``ctypes`` module is completely absent in Google AppEngine

            :seealso: https://github.com/cherrypy/cherrypy/issues/1506
            :seealso: https://github.com/cherrypy/cherrypy/issues/1512
            :ref: http://bit.ly/2gK6bXK
            """
            raise NotImplementedError
        else:
            return _argv

    @staticmethod
    def _extend_pythonpath(env):
        """Prepend current working dir to PATH environment variable if needed.

        If sys.path[0] is an empty string, the interpreter was likely
        invoked with -m and the effective path is about to change on
        re-exec.  Add the current directory to $PYTHONPATH to ensure
        that the new process sees the same path.

        This issue cannot be addressed in the general case because
        Python cannot reliably reconstruct the
        original command line (http://bugs.python.org/issue14208).

        (This idea filched from tornado.autoreload)
        """
        path_prefix = '.' + os.pathsep
        existing_path = env.get('PYTHONPATH', '')
        needs_patch = (
            sys.path[0] == '' and
            not existing_path.startswith(path_prefix)
        )

        if needs_patch:
            env['PYTHONPATH'] = path_prefix + existing_path

    def _set_cloexec(self):
        """Set the CLOEXEC flag on all open files (except stdin/out/err).

        If self.max_cloexec_files is an integer (the default), then on
        platforms which support it, it represents the max open files setting
        for the operating system. This function will be called just before
        the process is restarted via os.execv() to prevent open files
        from persisting into the new process.

        Set self.max_cloexec_files to 0 to disable this behavior.
        """
        for fd in range(3, self.max_cloexec_files):  # skip stdin/out/err
            try:
                flags = fcntl.fcntl(fd, fcntl.F_GETFD)
            except IOError:
                continue
            fcntl.fcntl(fd, fcntl.F_SETFD, flags | fcntl.FD_CLOEXEC)

    def stop(self):
        """Stop all services."""
        self.state = states.STOPPING
        self.log('Bus STOPPING')
        self.publish('stop')
        self.state = states.STOPPED
        self.log('Bus STOPPED')

    def start_with_callback(self, func, args=None, kwargs=None):
        """Start 'func' in a new thread T, then start self (and return T)."""
        if args is None:
            args = ()
        if kwargs is None:
            kwargs = {}
        args = (func,) + args

        def _callback(func, *a, **kw):
            self.wait(states.STARTED)
            func(*a, **kw)
        t = threading.Thread(target=_callback, args=args, kwargs=kwargs)
        t.setName('Bus Callback ' + t.getName())
        t.start()

        self.start()

        return t

    def log(self, msg='', level=20, traceback=False):
        """Log the given message. Append the last traceback if requested."""
        if traceback:
            msg += '\n' + ''.join(_traceback.format_exception(*sys.exc_info()))
        self.publish('log', msg, level)


bus = Bus()