##// END OF EJS Templates
wireproto: add streams to frame-based protocol...
wireproto: add streams to frame-based protocol Previously, the frame-based protocol was just a series of frames, with each frame associated with a request ID. In order to scale the protocol, we'll want to enable the use of compression. While it is possible to enable compression at the socket/pipe level, this has its disadvantages. The big one is it undermines the point of frames being standalone, atomic units that can be read and written: if you add compression above the framing protocol, you are back to having a stream-based protocol as opposed to something frame-based. So in order to preserve frames, compression needs to occur at the frame payload level. Compressing each frame's payload individually will limit compression ratios because the window size of the compressor will be limited by the max frame size, which is 32-64kb as currently defined. It will also add CPU overhead, as it is more efficient for compressors to operate on fewer, larger blocks of data than more, smaller blocks. So compressing each frame independently is out. This means we need to compress each frame's payload as if it is part of a larger stream. The simplest approach is to have 1 stream per connection. This could certainly work. However, it has disadvantages (documented below). We could also have 1 stream per RPC/command invocation. (This is the model HTTP/2 goes with.) This also has disadvantages. The main disadvantage to one global stream is that it has the very real potential to create CPU bottlenecks doing compression. Networks are only getting faster and the performance of single CPU cores has been relatively flat. Newer compression formats like zstandard offer better CPU cycle efficiency than predecessors like zlib. But it still all too common to saturate your CPU with compression overhead long before you saturate the network pipe. The main disadvantage with streams per request is that you can't reap the benefits of the compression context for multiple requests. For example, if you send 1000 RPC requests (or HTTP/2 requests for that matter), the response to each would have its own compression context. The overall size of the raw responses would be larger because compression contexts wouldn't be able to reference data from another request or response. The approach for streams as implemented in this commit is to support N streams per connection and for streams to potentially span requests and responses. As explained by the added internals docs, this facilitates servers and clients delegating independent streams and compression to independent threads / CPU cores. This helps alleviate the CPU bottleneck of compression. This design also allows compression contexts to be reused across requests/responses. This can result in improved compression ratios and less overhead for compressors and decompressors having to build new contexts. Another feature that was defined was the ability for individual frames within a stream to declare whether that individual frame's payload uses the content encoding (read: compression) defined by the stream. The idea here is that some servers may serve data from a combination of caches and dynamic resolution. Data coming from caches may be pre-compressed. We want to facilitate servers being able to essentially stream bytes from caches to the wire with minimal overhead. Being able to mix and match with frames are compressed within a stream enables these types of advanced server functionality. This commit defines the new streams mechanism. Basic code for supporting streams in frames has been added. But that code is seriously lacking and doesn't fully conform to the defined protocol. For example, we don't close any streams. And support for content encoding within streams is not yet implemented. The change was rather invasive and I didn't think it would be reasonable to implement the entire feature in a single commit. For the record, I would have loved to reuse an existing multiplexing protocol to build the new wire protocol on top of. However, I couldn't find a protocol that offers the performance and scaling characteristics that I desired. Namely, it should support multiple compression contexts to facilitate scaling out to multiple CPU cores and compression contexts should be able to live longer than single RPC requests. HTTP/2 *almost* fits the bill. But the semantics of HTTP message exchange state that streams can only live for a single request-response. We /could/ tunnel on top of HTTP/2 streams and frames with HEADER and DATA frames. But there's no guarantee that HTTP/2 libraries and proxies would allow us to use HTTP/2 streams and frames without the HTTP message exchange semantics defined in RFC 7540 Section 8. Other RPC protocols like gRPC tunnel are built on top of HTTP/2 and thus preserve its semantics of stream per RPC invocation. Even QUIC does this. We could attempt to invent a higher-level stream that spans HTTP/2 streams. But this would be violating HTTP/2 because there is no guarantee that HTTP/2 streams are routed to the same server. The best we can do - which is what this protocol does - is shoehorn all request and response data into a single HTTP message and create streams within. At that point, we've defined a Content-Type in HTTP parlance. It just so happens our media type can also work as a standalone, stream-based protocol, without leaning on HTTP or similar protocol. Differential Revision: https://phab.mercurial-scm.org/D2907

File last commit:

r35245:414114a7 default
r37304:9bfcbe4f default
Show More
selectors2.py
743 lines | 26.8 KiB | text/x-python | PythonLexer
""" Back-ported, durable, and portable selectors """
# MIT License
#
# Copyright (c) 2017 Seth Michael Larson
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from __future__ import absolute_import
import collections
import errno
import math
import select
import socket
import sys
import time
from .. import pycompat
namedtuple = collections.namedtuple
Mapping = collections.Mapping
try:
monotonic = time.monotonic
except AttributeError:
monotonic = time.time
__author__ = 'Seth Michael Larson'
__email__ = 'sethmichaellarson@protonmail.com'
__version__ = '2.0.0'
__license__ = 'MIT'
__url__ = 'https://www.github.com/SethMichaelLarson/selectors2'
__all__ = ['EVENT_READ',
'EVENT_WRITE',
'SelectorKey',
'DefaultSelector',
'BaseSelector']
EVENT_READ = (1 << 0)
EVENT_WRITE = (1 << 1)
_DEFAULT_SELECTOR = None
_SYSCALL_SENTINEL = object() # Sentinel in case a system call returns None.
_ERROR_TYPES = (OSError, IOError, socket.error)
SelectorKey = namedtuple('SelectorKey', ['fileobj', 'fd', 'events', 'data'])
class _SelectorMapping(Mapping):
""" Mapping of file objects to selector keys """
def __init__(self, selector):
self._selector = selector
def __len__(self):
return len(self._selector._fd_to_key)
def __getitem__(self, fileobj):
try:
fd = self._selector._fileobj_lookup(fileobj)
return self._selector._fd_to_key[fd]
except KeyError:
raise KeyError("{0!r} is not registered.".format(fileobj))
def __iter__(self):
return iter(self._selector._fd_to_key)
def _fileobj_to_fd(fileobj):
""" Return a file descriptor from a file object. If
given an integer will simply return that integer back. """
if isinstance(fileobj, int):
fd = fileobj
else:
try:
fd = int(fileobj.fileno())
except (AttributeError, TypeError, ValueError):
raise ValueError("Invalid file object: {0!r}".format(fileobj))
if fd < 0:
raise ValueError("Invalid file descriptor: {0}".format(fd))
return fd
class BaseSelector(object):
""" Abstract Selector class
A selector supports registering file objects to be monitored
for specific I/O events.
A file object is a file descriptor or any object with a
`fileno()` method. An arbitrary object can be attached to the
file object which can be used for example to store context info,
a callback, etc.
A selector can use various implementations (select(), poll(), epoll(),
and kqueue()) depending on the platform. The 'DefaultSelector' class uses
the most efficient implementation for the current platform.
"""
def __init__(self):
# Maps file descriptors to keys.
self._fd_to_key = {}
# Read-only mapping returned by get_map()
self._map = _SelectorMapping(self)
def _fileobj_lookup(self, fileobj):
""" Return a file descriptor from a file object.
This wraps _fileobj_to_fd() to do an exhaustive
search in case the object is invalid but we still
have it in our map. Used by unregister() so we can
unregister an object that was previously registered
even if it is closed. It is also used by _SelectorMapping
"""
try:
return _fileobj_to_fd(fileobj)
except ValueError:
# Search through all our mapped keys.
for key in self._fd_to_key.values():
if key.fileobj is fileobj:
return key.fd
# Raise ValueError after all.
raise
def register(self, fileobj, events, data=None):
""" Register a file object for a set of events to monitor. """
if (not events) or (events & ~(EVENT_READ | EVENT_WRITE)):
raise ValueError("Invalid events: {0!r}".format(events))
key = SelectorKey(fileobj, self._fileobj_lookup(fileobj), events, data)
if key.fd in self._fd_to_key:
raise KeyError("{0!r} (FD {1}) is already registered"
.format(fileobj, key.fd))
self._fd_to_key[key.fd] = key
return key
def unregister(self, fileobj):
""" Unregister a file object from being monitored. """
try:
key = self._fd_to_key.pop(self._fileobj_lookup(fileobj))
except KeyError:
raise KeyError("{0!r} is not registered".format(fileobj))
# Getting the fileno of a closed socket on Windows errors with EBADF.
except socket.error as err:
if err.errno != errno.EBADF:
raise
else:
for key in self._fd_to_key.values():
if key.fileobj is fileobj:
self._fd_to_key.pop(key.fd)
break
else:
raise KeyError("{0!r} is not registered".format(fileobj))
return key
def modify(self, fileobj, events, data=None):
""" Change a registered file object monitored events and data. """
# NOTE: Some subclasses optimize this operation even further.
try:
key = self._fd_to_key[self._fileobj_lookup(fileobj)]
except KeyError:
raise KeyError("{0!r} is not registered".format(fileobj))
if events != key.events:
self.unregister(fileobj)
key = self.register(fileobj, events, data)
elif data != key.data:
# Use a shortcut to update the data.
key = key._replace(data=data)
self._fd_to_key[key.fd] = key
return key
def select(self, timeout=None):
""" Perform the actual selection until some monitored file objects
are ready or the timeout expires. """
raise NotImplementedError()
def close(self):
""" Close the selector. This must be called to ensure that all
underlying resources are freed. """
self._fd_to_key.clear()
self._map = None
def get_key(self, fileobj):
""" Return the key associated with a registered file object. """
mapping = self.get_map()
if mapping is None:
raise RuntimeError("Selector is closed")
try:
return mapping[fileobj]
except KeyError:
raise KeyError("{0!r} is not registered".format(fileobj))
def get_map(self):
""" Return a mapping of file objects to selector keys """
return self._map
def _key_from_fd(self, fd):
""" Return the key associated to a given file descriptor
Return None if it is not found. """
try:
return self._fd_to_key[fd]
except KeyError:
return None
def __enter__(self):
return self
def __exit__(self, *_):
self.close()
# Almost all platforms have select.select()
if hasattr(select, "select"):
class SelectSelector(BaseSelector):
""" Select-based selector. """
def __init__(self):
super(SelectSelector, self).__init__()
self._readers = set()
self._writers = set()
def register(self, fileobj, events, data=None):
key = super(SelectSelector, self).register(fileobj, events, data)
if events & EVENT_READ:
self._readers.add(key.fd)
if events & EVENT_WRITE:
self._writers.add(key.fd)
return key
def unregister(self, fileobj):
key = super(SelectSelector, self).unregister(fileobj)
self._readers.discard(key.fd)
self._writers.discard(key.fd)
return key
def select(self, timeout=None):
# Selecting on empty lists on Windows errors out.
if not len(self._readers) and not len(self._writers):
return []
timeout = None if timeout is None else max(timeout, 0.0)
ready = []
r, w, _ = _syscall_wrapper(self._wrap_select, True, self._readers,
self._writers, timeout)
r = set(r)
w = set(w)
for fd in r | w:
events = 0
if fd in r:
events |= EVENT_READ
if fd in w:
events |= EVENT_WRITE
key = self._key_from_fd(fd)
if key:
ready.append((key, events & key.events))
return ready
def _wrap_select(self, r, w, timeout=None):
""" Wrapper for select.select because timeout is a positional arg """
return select.select(r, w, [], timeout)
__all__.append('SelectSelector')
# Jython has a different implementation of .fileno() for socket objects.
if pycompat.isjython:
class _JythonSelectorMapping(object):
""" This is an implementation of _SelectorMapping that is built
for use specifically with Jython, which does not provide a hashable
value from socket.socket.fileno(). """
def __init__(self, selector):
assert isinstance(selector, JythonSelectSelector)
self._selector = selector
def __len__(self):
return len(self._selector._sockets)
def __getitem__(self, fileobj):
for sock, key in self._selector._sockets:
if sock is fileobj:
return key
else:
raise KeyError("{0!r} is not registered.".format(fileobj))
class JythonSelectSelector(SelectSelector):
""" This is an implementation of SelectSelector that is for Jython
which works around that Jython's socket.socket.fileno() does not
return an integer fd value. All SelectorKey.fd will be equal to -1
and should not be used. This instead uses object id to compare fileobj
and will only use select.select as it's the only selector that allows
directly passing in socket objects rather than registering fds.
See: http://bugs.jython.org/issue1678
https://wiki.python.org/jython/NewSocketModule#socket.fileno.28.29_does_not_return_an_integer
"""
def __init__(self):
super(JythonSelectSelector, self).__init__()
self._sockets = [] # Uses a list of tuples instead of dictionary.
self._map = _JythonSelectorMapping(self)
self._readers = []
self._writers = []
# Jython has a select.cpython_compatible_select function in older versions.
self._select_func = getattr(select, 'cpython_compatible_select', select.select)
def register(self, fileobj, events, data=None):
for sock, _ in self._sockets:
if sock is fileobj:
raise KeyError("{0!r} is already registered"
.format(fileobj, sock))
key = SelectorKey(fileobj, -1, events, data)
self._sockets.append((fileobj, key))
if events & EVENT_READ:
self._readers.append(fileobj)
if events & EVENT_WRITE:
self._writers.append(fileobj)
return key
def unregister(self, fileobj):
for i, (sock, key) in enumerate(self._sockets):
if sock is fileobj:
break
else:
raise KeyError("{0!r} is not registered.".format(fileobj))
if key.events & EVENT_READ:
self._readers.remove(fileobj)
if key.events & EVENT_WRITE:
self._writers.remove(fileobj)
del self._sockets[i]
return key
def _wrap_select(self, r, w, timeout=None):
""" Wrapper for select.select because timeout is a positional arg """
return self._select_func(r, w, [], timeout)
__all__.append('JythonSelectSelector')
SelectSelector = JythonSelectSelector # Override so the wrong selector isn't used.
if hasattr(select, "poll"):
class PollSelector(BaseSelector):
""" Poll-based selector """
def __init__(self):
super(PollSelector, self).__init__()
self._poll = select.poll()
def register(self, fileobj, events, data=None):
key = super(PollSelector, self).register(fileobj, events, data)
event_mask = 0
if events & EVENT_READ:
event_mask |= select.POLLIN
if events & EVENT_WRITE:
event_mask |= select.POLLOUT
self._poll.register(key.fd, event_mask)
return key
def unregister(self, fileobj):
key = super(PollSelector, self).unregister(fileobj)
self._poll.unregister(key.fd)
return key
def _wrap_poll(self, timeout=None):
""" Wrapper function for select.poll.poll() so that
_syscall_wrapper can work with only seconds. """
if timeout is not None:
if timeout <= 0:
timeout = 0
else:
# select.poll.poll() has a resolution of 1 millisecond,
# round away from zero to wait *at least* timeout seconds.
timeout = math.ceil(timeout * 1000)
result = self._poll.poll(timeout)
return result
def select(self, timeout=None):
ready = []
fd_events = _syscall_wrapper(self._wrap_poll, True, timeout=timeout)
for fd, event_mask in fd_events:
events = 0
if event_mask & ~select.POLLIN:
events |= EVENT_WRITE
if event_mask & ~select.POLLOUT:
events |= EVENT_READ
key = self._key_from_fd(fd)
if key:
ready.append((key, events & key.events))
return ready
__all__.append('PollSelector')
if hasattr(select, "epoll"):
class EpollSelector(BaseSelector):
""" Epoll-based selector """
def __init__(self):
super(EpollSelector, self).__init__()
self._epoll = select.epoll()
def fileno(self):
return self._epoll.fileno()
def register(self, fileobj, events, data=None):
key = super(EpollSelector, self).register(fileobj, events, data)
events_mask = 0
if events & EVENT_READ:
events_mask |= select.EPOLLIN
if events & EVENT_WRITE:
events_mask |= select.EPOLLOUT
_syscall_wrapper(self._epoll.register, False, key.fd, events_mask)
return key
def unregister(self, fileobj):
key = super(EpollSelector, self).unregister(fileobj)
try:
_syscall_wrapper(self._epoll.unregister, False, key.fd)
except _ERROR_TYPES:
# This can occur when the fd was closed since registry.
pass
return key
def select(self, timeout=None):
if timeout is not None:
if timeout <= 0:
timeout = 0.0
else:
# select.epoll.poll() has a resolution of 1 millisecond
# but luckily takes seconds so we don't need a wrapper
# like PollSelector. Just for better rounding.
timeout = math.ceil(timeout * 1000) * 0.001
timeout = float(timeout)
else:
timeout = -1.0 # epoll.poll() must have a float.
# We always want at least 1 to ensure that select can be called
# with no file descriptors registered. Otherwise will fail.
max_events = max(len(self._fd_to_key), 1)
ready = []
fd_events = _syscall_wrapper(self._epoll.poll, True,
timeout=timeout,
maxevents=max_events)
for fd, event_mask in fd_events:
events = 0
if event_mask & ~select.EPOLLIN:
events |= EVENT_WRITE
if event_mask & ~select.EPOLLOUT:
events |= EVENT_READ
key = self._key_from_fd(fd)
if key:
ready.append((key, events & key.events))
return ready
def close(self):
self._epoll.close()
super(EpollSelector, self).close()
__all__.append('EpollSelector')
if hasattr(select, "devpoll"):
class DevpollSelector(BaseSelector):
"""Solaris /dev/poll selector."""
def __init__(self):
super(DevpollSelector, self).__init__()
self._devpoll = select.devpoll()
def fileno(self):
return self._devpoll.fileno()
def register(self, fileobj, events, data=None):
key = super(DevpollSelector, self).register(fileobj, events, data)
poll_events = 0
if events & EVENT_READ:
poll_events |= select.POLLIN
if events & EVENT_WRITE:
poll_events |= select.POLLOUT
self._devpoll.register(key.fd, poll_events)
return key
def unregister(self, fileobj):
key = super(DevpollSelector, self).unregister(fileobj)
self._devpoll.unregister(key.fd)
return key
def _wrap_poll(self, timeout=None):
""" Wrapper function for select.poll.poll() so that
_syscall_wrapper can work with only seconds. """
if timeout is not None:
if timeout <= 0:
timeout = 0
else:
# select.devpoll.poll() has a resolution of 1 millisecond,
# round away from zero to wait *at least* timeout seconds.
timeout = math.ceil(timeout * 1000)
result = self._devpoll.poll(timeout)
return result
def select(self, timeout=None):
ready = []
fd_events = _syscall_wrapper(self._wrap_poll, True, timeout=timeout)
for fd, event_mask in fd_events:
events = 0
if event_mask & ~select.POLLIN:
events |= EVENT_WRITE
if event_mask & ~select.POLLOUT:
events |= EVENT_READ
key = self._key_from_fd(fd)
if key:
ready.append((key, events & key.events))
return ready
def close(self):
self._devpoll.close()
super(DevpollSelector, self).close()
__all__.append('DevpollSelector')
if hasattr(select, "kqueue"):
class KqueueSelector(BaseSelector):
""" Kqueue / Kevent-based selector """
def __init__(self):
super(KqueueSelector, self).__init__()
self._kqueue = select.kqueue()
def fileno(self):
return self._kqueue.fileno()
def register(self, fileobj, events, data=None):
key = super(KqueueSelector, self).register(fileobj, events, data)
if events & EVENT_READ:
kevent = select.kevent(key.fd,
select.KQ_FILTER_READ,
select.KQ_EV_ADD)
_syscall_wrapper(self._kqueue.control, False, [kevent], 0, 0)
if events & EVENT_WRITE:
kevent = select.kevent(key.fd,
select.KQ_FILTER_WRITE,
select.KQ_EV_ADD)
_syscall_wrapper(self._kqueue.control, False, [kevent], 0, 0)
return key
def unregister(self, fileobj):
key = super(KqueueSelector, self).unregister(fileobj)
if key.events & EVENT_READ:
kevent = select.kevent(key.fd,
select.KQ_FILTER_READ,
select.KQ_EV_DELETE)
try:
_syscall_wrapper(self._kqueue.control, False, [kevent], 0, 0)
except _ERROR_TYPES:
pass
if key.events & EVENT_WRITE:
kevent = select.kevent(key.fd,
select.KQ_FILTER_WRITE,
select.KQ_EV_DELETE)
try:
_syscall_wrapper(self._kqueue.control, False, [kevent], 0, 0)
except _ERROR_TYPES:
pass
return key
def select(self, timeout=None):
if timeout is not None:
timeout = max(timeout, 0)
max_events = len(self._fd_to_key) * 2
ready_fds = {}
kevent_list = _syscall_wrapper(self._kqueue.control, True,
None, max_events, timeout)
for kevent in kevent_list:
fd = kevent.ident
event_mask = kevent.filter
events = 0
if event_mask == select.KQ_FILTER_READ:
events |= EVENT_READ
if event_mask == select.KQ_FILTER_WRITE:
events |= EVENT_WRITE
key = self._key_from_fd(fd)
if key:
if key.fd not in ready_fds:
ready_fds[key.fd] = (key, events & key.events)
else:
old_events = ready_fds[key.fd][1]
ready_fds[key.fd] = (key, (events | old_events) & key.events)
return list(ready_fds.values())
def close(self):
self._kqueue.close()
super(KqueueSelector, self).close()
__all__.append('KqueueSelector')
def _can_allocate(struct):
""" Checks that select structs can be allocated by the underlying
operating system, not just advertised by the select module. We don't
check select() because we'll be hopeful that most platforms that
don't have it available will not advertise it. (ie: GAE) """
try:
# select.poll() objects won't fail until used.
if struct == 'poll':
p = select.poll()
p.poll(0)
# All others will fail on allocation.
else:
getattr(select, struct)().close()
return True
except (OSError, AttributeError):
return False
# Python 3.5 uses a more direct route to wrap system calls to increase speed.
if sys.version_info >= (3, 5):
def _syscall_wrapper(func, _, *args, **kwargs):
""" This is the short-circuit version of the below logic
because in Python 3.5+ all selectors restart system calls. """
return func(*args, **kwargs)
else:
def _syscall_wrapper(func, recalc_timeout, *args, **kwargs):
""" Wrapper function for syscalls that could fail due to EINTR.
All functions should be retried if there is time left in the timeout
in accordance with PEP 475. """
timeout = kwargs.get("timeout", None)
if timeout is None:
expires = None
recalc_timeout = False
else:
timeout = float(timeout)
if timeout < 0.0: # Timeout less than 0 treated as no timeout.
expires = None
else:
expires = monotonic() + timeout
args = list(args)
if recalc_timeout and "timeout" not in kwargs:
raise ValueError(
"Timeout must be in args or kwargs to be recalculated")
result = _SYSCALL_SENTINEL
while result is _SYSCALL_SENTINEL:
try:
result = func(*args, **kwargs)
# OSError is thrown by select.select
# IOError is thrown by select.epoll.poll
# select.error is thrown by select.poll.poll
# Aren't we thankful for Python 3.x rework for exceptions?
except (OSError, IOError, select.error) as e:
# select.error wasn't a subclass of OSError in the past.
errcode = None
if hasattr(e, "errno"):
errcode = e.errno
elif hasattr(e, "args"):
errcode = e.args[0]
# Also test for the Windows equivalent of EINTR.
is_interrupt = (errcode == errno.EINTR or (hasattr(errno, "WSAEINTR") and
errcode == errno.WSAEINTR))
if is_interrupt:
if expires is not None:
current_time = monotonic()
if current_time > expires:
raise OSError(errno=errno.ETIMEDOUT)
if recalc_timeout:
if "timeout" in kwargs:
kwargs["timeout"] = expires - current_time
continue
raise
return result
# Choose the best implementation, roughly:
# kqueue == devpoll == epoll > poll > select
# select() also can't accept a FD > FD_SETSIZE (usually around 1024)
def DefaultSelector():
""" This function serves as a first call for DefaultSelector to
detect if the select module is being monkey-patched incorrectly
by eventlet, greenlet, and preserve proper behavior. """
global _DEFAULT_SELECTOR
if _DEFAULT_SELECTOR is None:
if pycompat.isjython:
_DEFAULT_SELECTOR = JythonSelectSelector
elif _can_allocate('kqueue'):
_DEFAULT_SELECTOR = KqueueSelector
elif _can_allocate('devpoll'):
_DEFAULT_SELECTOR = DevpollSelector
elif _can_allocate('epoll'):
_DEFAULT_SELECTOR = EpollSelector
elif _can_allocate('poll'):
_DEFAULT_SELECTOR = PollSelector
elif hasattr(select, 'select'):
_DEFAULT_SELECTOR = SelectSelector
else: # Platform-specific: AppEngine
raise RuntimeError('Platform does not have a selector.')
return _DEFAULT_SELECTOR()