Source

gipc / gipc / gipc.py

Full commit
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
# -*- coding: utf-8 -*-
#
#   Copyright (C) 2012 Jan-Philip Gehrcke
#
#   Licensed under the Apache License, Version 2.0 (the "License");
#   you may not use this file except in compliance with the License.
#   You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
#   Unless required by applicable law or agreed to in writing, software
#   distributed under the License is distributed on an "AS IS" BASIS,
#   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#   See the License for the specific language governing permissions and
#   limitations under the License.

"""
.. todo::

    - Implement poll/peek on read end. (It's impossible to identify complete
        messages in advance, but within the framework only complete messages
        are sent.)
    - Can gevent's FileObjectPosix be of any use?
    - Review buffer-implementation, consider, buffer(), memoryview(), ..
    - hub.cancel_wait() (cf. gevent sockets) in close instead of lock check?
    - Ensure portability between Python 2 and 3
    - put() timeout? Relevant in case of filled up pipes. However, put()
        duration cannot be controlled if write is blocking *after* partial
        msg write. Doesn't make sense I think.

    - pipe implementation on Windows based on NamedPipes with overlapping IO
        could give useful control:
        - async IO, peeknamedpipe, ReadfileEx, SetCommTimeouts, CancelIO
        - use pywin32 or http://docs.python.org/2/library/ctypes.html
        - problem: circumvent blocking Win32 calls (hand control to gevent)
        - async IO without blocking any thread: IOCP
          http://msdn.microsoft.com/en-us/library/aa365198%28VS.85%29.aspx
"""

import os
import sys
import io
import struct
import itertools
import signal
import logging
import multiprocessing
import multiprocessing.forking
import multiprocessing.process
try:
   import cPickle as pickle
except:
   import pickle
WINDOWS = sys.platform == "win32"
if WINDOWS:
    import msvcrt
import gevent
import gevent.os
import gevent.lock
import gevent.event


log = logging.getLogger("gipc")


class GIPCError(Exception):
    """Is raised upon general errors. All other exception types derive from
    this one.
    """
    pass


class GIPCClosed(GIPCError):
    """Is raised upon operation on closed handle.
    """
    pass


class GIPCLocked(GIPCError):
    """Is raised upon attempt to close a handle which is currently locked for
    I/O.
    """
    pass


def pipe():
    """Creates a new pipe and returns its corresponding read and write
    handles. Those allow for sending and receiving pickleable objects through
    the pipe in a gevent-cooperative manner. A handle-pair can transmit
    data between greenlets within one process or across processes created via
    :func:`start_process`.

    :returns:
        ``(reader, writer)`` tuple. Both items are instances of
        :class:`gipc._GIPCHandle`.

    :class:`gipc._GIPCHandle` instances are recommended to be used with Python's
    context manager in the following ways::

        with pipe() as (r, w):
            do_something(r, w)

    ::

        reader, writer = pipe()

        with reader:
            do_something(reader)

        with writer as w:
            do_something(w)

    The transport layer is based on ``os.pipe()`` (i.e. ``CreatePipe()`` on
    Windows and ``pipe()`` on POSIX-compliant systems).
    """
    # os.pipe() implementation on Windows (http://bit.ly/RDuKUm):
    #   - CreatePipe(&read, &write, NULL, 0)
    #   - anonymous pipe, system handles buffer size
    #   - anonymous pipes are implemented using named pipes with unique names
    #   - asynchronous (overlapped) read and write operations not supported
    # os.pipe() implementation on Unix (http://linux.die.net/man/7/pipe):
    #   - based on pipe()
    #   - common Linux: pipe buffer is 4096 bytes, pipe capacity is 65536 bytes
    r, w = os.pipe()
    reader = _GIPCReader(r)
    writer = _GIPCWriter(w)
    _all_handles.append(reader)
    _all_handles.append(writer)
    return _HandlePairContext((reader, writer))


def start_process(target, args=(), kwargs={}, daemon=None, name=None):
    """Spawn child process and execute function ``target(*args, **kwargs)``.
    Any existing :class:`gipc._GIPCHandle` can be handed over to the child process
    via ``args`` and/or ``kwargs``.

    :arg target:
        Function to be called in child as ``target(*args, **kwargs)``.

    :arg args:
        Tuple defining positional arguments provided to ``target``.

    :arg kwargs:
        Dictionary defining keyword arguments provided to ``target``.

    :arg name:
        Forwarded to ``multiprocessing.Process.name``.

    :arg daemon:
        Forwarded to ``multiprocessing.Process.daemon``.

    :returns:
        :class:`gipc._GProcess` instance (inherits from
        ``multiprocessing.Process`` and re-implements some of its methods in a
        gevent-cooperative fashion).

    Process creation is based on ``multiprocessing.Process()``. When working
    with gevent, instead of calling ``Process()`` directly, it is highly
    recommended to create child processes via :func:`start_process`. It takes
    care of

        - closing dispensable file descriptors after child process creation.
        - proper file descriptor inheritance on Windows.
        - re-initialization of the event loop in the child process.
        - providing cooperative process methods (such as ``join()``).

    Calling this method breaks ``os.waitpid()`` on Unix: spawning the first
    child activates libev child watchers, leading to libev reap children in the
    moment they die. Applied to such a child, ``os.waitpid()`` throws
    ``ECHILD`` (cf. http://linux.die.net/man/2/waitid).
    """
    if not isinstance(args, tuple):
        raise TypeError, '`args` must be tuple.'
    if not isinstance(kwargs, dict):
        raise TypeError, '`kwargs` must be dictionary.'
    log.debug("Invoke target `%s` in child process." % target)
    allargs = itertools.chain(args, kwargs.values())
    childhandles = [a for a in allargs if isinstance(a, _GIPCHandle)]
    if WINDOWS:
        for h in _all_handles:
            h._pre_createprocess_windows()
    p = _GProcess(
        target=_child,
        name=name,
        kwargs={"target": target,
                "all_handles": _all_handles,
                "args": args,
                "kwargs": kwargs})
    if daemon is not None:
        p.daemon = daemon
    p.start()
    if WINDOWS:
        for h in _all_handles:
            h._post_createprocess_windows()
    # Close dispensable file handles in parent.
    for h in childhandles:
        log.debug("Invalidate %s in parent." % h)
        h.close()
    return p


def _child(target, all_handles, args, kwargs):
    """Wrapper function that runs in child process. Sanitizes situation and
    executes user-given function.

    After fork on POSIX-compliant systems, gevent's state is inherited by the
    child which may lead to undesired behavior, such as greenlets running in
    both, the parent and the child. Therefore, on Unix, gevent's state is
    entirely reset before running the user-given function.
    """
    log.debug("_child start. target: `%s`" % target)
    # Restore global `_all_handles` (required on Win, does not harm elsewhere).
    # (The value of a global variable set in the parent process is not
    #  propagated to children on Windows (as is on Unix via fork())).
    global _all_handles
    _all_handles = all_handles
    if not WINDOWS:
        # `gevent.reinit` calls `libev.ev_loop_fork()`, which is designed to
        # be called after fork.
        gevent.reinit()
        # Destroy default event loop via `libev.ev_loop_destroy()` and delete
        # hub. This dumps all registered events and greenlets that have been
        # duplicated from the parent via fork().
        gevent.get_hub().destroy(destroy_loop=True)
        # Create a new hub and a new default event loop via
        # `libev.gevent_ev_default_loop`.
        h = gevent.get_hub(default=True)
        assert h.loop.default, 'Could not create libev default event loop.'
    allargs = itertools.chain(args, kwargs.values())
    childhandles = [a for a in allargs if isinstance(a, _GIPCHandle)]
    # Register inherited handles for current process. Close those that are not
    # intended for further usage.
    for h in _all_handles[:]:
        h._set_legit_process()
        if WINDOWS:
            h._post_createprocess_windows()
        if not h in childhandles:
            log.debug("Invalidate %s in child." % h)
            h.close()
            continue
        log.debug("Handle `%s` is valid in child." % h)
    target(*args, **kwargs)
    for h in childhandles:
        try:
            # The user or a child of this child might already have closen it.
            h.close()
        except GIPCClosed:
            pass


class _GProcess(multiprocessing.Process):
    """
    Implements adjustments to multiprocessing's Process class for
    gevent-cooperativity. Currently re-implements ``start()``, ``is_alive()``,
    ``exitcode`` on Unix and ``join()`` on Windows and Unix.
    """
    #On Unix, we  cannot rely on `multiprocessing.Process.is_alive()` and
    #`multiprocessing.Process._popen.wait()` to tell the truth about the state
    #of children of children:

    #In the initial process, gevent makes libev's default event loop not
    #reap dead children. In children, however, after initialization of the
    #libev default event loop, dead children (grandchildren of the initial
    #process) become reaped immediately by the event loop.

    #This makes `os.waitpid(grandchild_pid)` throw an ECHILD error (process
    #specified by pid does not exist or is not a child of the calling process).
    #This leads to multiprocessing's `_popen.wait()` returning `None`, meaning
    #'alive' -- for child processes that does not exist anymore.

    #Immediate child reaping by libev could be rectified via re-installing
    #the default signal handler with `signal(signal.SIGCHLD, signal.SIG_DFL)`.
    #However, doing so would render libev's child watchers useless.

    #Instead, for each `_GProcess`, a libev child watcher is explicitly
    #started in the modified `start()` method below. The modified `join()`
    #method is adjusted to this libev-watcher-based child monitoring.
    #`multiprocessing.Process.join()` is entirely surpassed, but resembled.
    #`os.waitpid()` is broken in all process generations after the first
    #`_GProcess` has been started.

    #On Windows, cooperative `join()` is realized via frequent non-blocking
    #calls to `Process.is_alive()` and the original `join()` method.
    if not WINDOWS:
        def start(self):
            hub = gevent.get_hub()
            self._returnevent = gevent.event.Event()
            # Start grabbing SIGCHLD in libev event loop.
            hub.loop.install_sigchld()
            # Run new process (based on `fork()` on POSIX-compliant systems,
            # and on `CreateProcess()` on Windows).
            super(_GProcess, self).start()
            # The occurrence of SIGCHLD is recorded asynchronously in libev.
            # This guarantees proper behaviour even if the child watcher is
            # started after the child exits. Start child watcher now.
            self._sigchld_watcher = hub.loop.child(self.pid)
            self._sigchld_watcher.start(
                self._on_sigchld, self._sigchld_watcher)
            log.debug("SIGCHLD watcher for %s started." % self.pid)

        def _on_sigchld(self, watcher):
            watcher.stop()
            # Status evaluation taken from multiprocessing.forking.
            if os.WIFSIGNALED(watcher.rstatus):
                self._popen.returncode = -os.WTERMSIG(watcher.rstatus)
            else:
                assert os.WIFEXITED(watcher.rstatus)
                self._popen.returncode = os.WEXITSTATUS(watcher.rstatus)
            self._returnevent.set()
            log.debug(("SIGCHLD watcher callback for %s invoked. Exitcode "
                       "stored: %s" % (self.pid, self._popen.returncode)))

        def is_alive(self):
            if self._popen.returncode is None:
                return True
            return False

        @property
        def exitcode(self):
            return self._popen.returncode

    def join(self, timeout=None):
        """
        Wait cooperatively until child process terminates or timeout occurs.

        :arg timeout: ``None`` (default) or a a time in seconds. The method
            simply returns upon timeout expiration. The state of the process
            has to be identified via ``is_alive()``.
        """
        assert self._parent_pid == os.getpid(), "I'm not parent of this child."
        assert self._popen is not None, 'Can only join a started process.'
        if not WINDOWS:
            # Resemble multiprocessing's join() method while replacing
            # `self._popen.wait(timeout)` with
            # `self._returnevent.wait(timeout)`
            self._returnevent.wait(timeout)
            if self._popen.returncode is not None:
                multiprocessing.process._current_process._children.discard(
                    self)
            return
        with gevent.Timeout(timeout, False):
            while self.is_alive():
                # Is the polling frequency reasonable?
                gevent.sleep(0.01)
        # Clean up after child as designed by Process class (non-blocking).
        super(_GProcess, self).join(timeout=0)


class _GIPCHandle(object):
    """
    The ``_GIPCHandle`` class implements common features of read and write
    handles. ``_GIPCHandle`` instances are created via :func:`pipe`.

    .. todo::

        Care about destructor?
        http://eli.thegreenplace.net/2009/06/12/
        safely-using-destructors-in-python/
    """
    def __init__(self):
        self._id = os.urandom(3).encode("hex")
        self._legit_pid = os.getpid()
        self._make_nonblocking()
        self._lock = gevent.lock.Semaphore(value=1)
        self._closed = False

    def _make_nonblocking(self):
        if hasattr(gevent.os, 'make_nonblocking'):
            # On POSIX-compliant systems, the file descriptor flags are
            # inherited after forking, i.e. it is sufficient to make fd
            # nonblocking only once.
            gevent.os.make_nonblocking(self._fd)

    def close(self):
        """Close underlying file descriptor and de-register handle for further
        usage. Is called on context exit.

        Raises:
            - :exc:`GIPCError`
            - :exc:`GIPCClosed`
            - :exc:`GIPCLocked`
        """
        global _all_handles
        self._validate()
        if not self._lock.acquire(blocking=False):
            raise GIPCLocked(
                "Can't close handle %s: locked for I/O operation." % self)
        log.debug("Invalidating %s ..." % self)
        self._closed = True
        if self._fd is not None:
            os.close(self._fd)
            self._fd = None
        if self in _all_handles:
            # Remove the handle from the global list of valid handles.
            _all_handles.remove(self)
        self._lock.release()

    def _set_legit_process(self):
        log.debug("Legitimate %s for current process." % self)
        self._legit_pid = os.getpid()

    def _validate(self):
        """Raise exception if this handle is closed or not registered to be
        used in the current process.

        Intended to be called before every operation on `self._fd`.
        Reveals wrong usage of this module in the context of multiple
        processes. Might prevent tedious debugging sessions. Has little
        performance impact.
        """
        if self._closed:
            raise GIPCClosed(
                "GIPCHandle has been closed before.")
        if os.getpid() != self._legit_pid:
            raise GIPCError(
                "GIPCHandle %s not registered for current process %s." % (
                self, os.getpid()))

    def _pre_createprocess_windows(self):
        """Prepare file descriptor for transfer to child process on Windows.

        By default, file descriptors are not inherited by child processes on
        Windows. However, they can be made inheritable via calling the system
        function `DuplicateHandle` while setting `bInheritHandle` to True.
        From MSDN:
            bInheritHandle:
                A variable that indicates whether the handle is inheritable.
                If TRUE, the duplicate handle can be inherited by new processes
                created by the target process. If FALSE, the new handle cannot
                be inherited.
        The Python `subprocess` and `multiprocessing` modules make use of this.
        There is no Python API officially exposed. However, the function
        `multiprocessing.forking.duplicate` is available since the introduction
        of the multiprocessing module in Python 2.6 up to the development
        version of Python 3.4 as of 2012-10-20.
        """
        if WINDOWS:
            from multiprocessing.forking import duplicate
            # Get Windows file handle from C file descriptor.
            h = msvcrt.get_osfhandle(self._fd)
            # Duplicate file handle, rendering the duplicate inheritable by
            # processes created by the current process.
            self._ihfd = duplicate(handle=h, inheritable=True)
            # Close "old" (in-inheritable) file descriptor.
            os.close(self._fd)
            self._fd = False

    def _post_createprocess_windows(self):
        """Restore file descriptor on Windows."""
        if WINDOWS:
            # Get C file descriptor from Windows file handle.
            self._fd = msvcrt.open_osfhandle(self._ihfd, self._fd_flag)
            del self._ihfd

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_value, traceback):
        try:
            self.close()
        except GIPCClosed:
            # Closed before, which is fine.
            pass
        except GIPCLocked:
            # Locked for I/O outside of context, which is not fine.
            raise GIPCLocked((
                "Context manager can't close handle %s. It's locked for I/O "
                "operation out of context." % self))

    def __str__(self):
        return self.__repr__()

    def __repr__(self):
        fd = self._fd
        if hasattr(self, "_ihfd"):
            fd = "WIN_%s" % self._ihfd
        return "<%s_%s fd: %s>" % (self.__class__.__name__, self._id, fd)


class _GIPCReader(_GIPCHandle):
    """
    A ``_GIPCReader`` instance manages the read end of a pipe. It is created
    via :func:`pipe`.
    """
    def __init__(self, pipe_read_fd):
        self._fd = pipe_read_fd
        self._fd_flag = os.O_RDONLY
        _GIPCHandle.__init__(self)
        self._newmessage_lengthreceived = False
        self._timeout = None

    def _recv_in_buffer(self, n):
        """Cooperatively read `n` bytes from file descriptor to buffer."""
        readbuf = io.BytesIO()
        remaining = n
        while remaining > 0:
            chunk = _READ_NB(self._fd, remaining)
            received = len(chunk)
            if received == 0:
                if remaining == n:
                    raise EOFError(
                        "Most likely, the other pipe end is closed.")
                else:
                    raise IOError("Message interrupted by EOF.")
            readbuf.write(chunk)
            remaining -= received
        return readbuf

    def get(self, timeout=None):
        """Receive and return an object from the pipe. Block
        gevent-cooperatively until object is available or timeout expires.

        :arg timeout: ``None`` (default) or a ``gevent.Timeout``
            instance. The timeout must be started to take effect and is
            cancelled when the first byte of a new message arrives (i.e.
            providing a timeout does not guarantee that the method completes
            within the timeout interval).

        :returns: a Python object.

        Raises:
            - :exc:`gevent.Timeout` (if provided)
            - :exc:`GIPCError`
            - :exc:`GIPCClosed`
            - :exc:`pickle.UnpicklingError`

        Recommended usage for silent timeout control::

            with gevent.Timeout(TIME_SECONDS, False) as t:
                reader.get(timeout=t)

        .. warning::

            The timeout control is currently not available on Windows,
            because Windows can't apply select() to pipe handles.
            An ``OSError`` is expected to be raised in case you set a
            timeout.
        """
        self._validate()
        with self._lock:
            if timeout:
                # Wait for ready-to-read event.
                h = gevent.get_hub()
                h.wait(h.loop.io(self._fd, 1))
                timeout.cancel()
            msize, = struct.unpack("!i", self._recv_in_buffer(4).getvalue())
            self._newmessage_lengthreceived = True
            bindata = self._recv_in_buffer(msize).getvalue()
            self._newmessage_lengthreceived = False
        return pickle.loads(bindata)


class _GIPCWriter(_GIPCHandle):
    """
    A ``_GIPCWriter`` instance manages the write end of a pipe. It is created
    via :func:`pipe`.
    """
    def __init__(self, pipe_write_fd):
        self._fd = pipe_write_fd
        self._fd_flag = os.O_WRONLY
        _GIPCHandle.__init__(self)

    def _write(self, bindata):
        """Write `bindata` to pipe in a gevent-cooperative manner.

        POSIX-compliant system notes (http://linux.die.net/man/7/pipe:):
            - Since Linux 2.6.11, the pipe capacity is 65536 bytes
            - Relevant for large messages (O_NONBLOCK enabled,
              n > PIPE_BUF (4096 Byte, usually)):
                "If the pipe is full, then write(2) fails, with errno set
                to EAGAIN. Otherwise, from 1 to n bytes may be written (i.e.,
                a "partial write" may occur; the caller should check the
                return value from write(2) to see how many bytes were
                actualy written), and these bytes may be interleaved with
                writes by other processes."

            EAGAIN is handled within _WRITE_NB; partial writes here.
        """
        while True:
            # Causes OSError when read end is closed (broken pipe).
            diff = len(bindata) - _WRITE_NB(self._fd, bindata)
            if not diff:
                break
            bindata = bindata[-diff:]

    def put(self, o):
        """Pickle object ``o`` and write it to the pipe. Block
        gevent-cooperatively until all data is written.

        :arg o: a pickleable Python object.

        Raises:
            - :exc:`GIPCError`
            - :exc:`GIPCClosed`
            - :exc:`pickle.PicklingError`

        """
        self._validate()
        with self._lock:
            bindata = pickle.dumps(o, pickle.HIGHEST_PROTOCOL)
            # TODO: one write instead of two?
            self._write(struct.pack("!i", len(bindata)))
            self._write(bindata)


class _HandlePairContext(tuple):
    def __init__(self, (reader, writer)):
        self.reader = reader
        self.writer = writer
        super(_HandlePairContext, self).__init__((reader, writer))

    def __enter__(self):
        for h in self:
            h.__enter__()
        return self

    def __exit__(self, exc_type, exc_value, traceback):
        """
        Call `__exit__()` for both, read and write handles, in any case,
        as expected of a context manager. Exit writer first, as `os.close()`
        on reader might block on Windows otherwise. If an exception occurs
        during writer exit, store it, exit reader and raise it afterwards. If
        an exception is raised during both, reader and writer exit, only
        raise the reader exit exception.
        """
        writer_exit_exception = None
        try:
            self.writer.__exit__(exc_type, exc_value, traceback)
        except:
            writer_exit_exception = sys.exc_info()
        self.reader.__exit__(exc_type, exc_value, traceback)
        if writer_exit_exception:
            raise writer_exit_exception[1], None, writer_exit_exception[2]


# Define non-blocking read and write functions
if hasattr(gevent.os, 'nb_write'):
    # POSIX system -> use actual non-blocking I/O
    _READ_NB = gevent.os.nb_read
    _WRITE_NB = gevent.os.nb_write
else:
    # Windows -> imitate non-blocking I/O based on gevent threadpool
    _READ_NB = gevent.os.tp_read
    _WRITE_NB = gevent.os.tp_write


# Define container for keeping track of valid `_GIPCHandle`s.
_all_handles = []


def get_all_handles():
    return _all_handles[:]


def set_all_handles(handles):
    global _all_handles
    _all_handles = handles