[PATCH 1 of 6] socketserver: backport a less sucky socketserver from python 3.5.1
Jun Wu
quark at fb.com
Mon May 9 00:06:20 UTC 2016
# HG changeset patch
# User Jun Wu <quark at fb.com>
# Date 1462472424 -3600
# Thu May 05 19:20:24 2016 +0100
# Node ID fff86624fbfa0c2e78bb2874d68e6dd191c371cd
# Parent 57a24a85a2bc484cbd7c4b80959dc2d5f0483174
socketserver: backport a less sucky socketserver from python 3.5.1
The SocketServer shipped with Python 2.6 and 2.7 have many issues:
- Critical race condition (Python 2.6)
A race condition in ForkingMixIn that can lead to server hang forever with
100% CPU. See https://bugs.python.org/issue21491.
- Zombie processes (Python 2.6, 2.7)
ForkingMixIn leaves zombie processes. See https://bugs.python.org/issue11109.
It can be worked around by a custom SIGCHLD signal handler.
- Forced to use threads for periodic tasks (Python 2.6, 2.7, 3.x [1])
The text in serve_forever method suggests using threads for periodic tasks.
For chgserver, it has very lightweight checks on the socket file and time.
Being forced to use a thread sounds bad.
[1]: Note: on Python 3.x, the text remains but it's actually possible to
override service_actions to do periodic tasks.
- No modern object inheritance (Python 2.6, 2.7, 3.x)
The base classes and mix-ins use traditional objects so we cannot use "super"
in our own mix-ins and have to figure out the parent class carefully.
This is tolerable compared with the issues above.
Consider the facts that:
- We need to support Python 2.6 now
- There are other users of SocketServer: commandserver, hgweb. hgweb can be
affected by the Python 2.6 race condition if it uses the forking model.
It is okay to re-invent the logic for chgserver alone but supporting all
existing code is more complex and does not worth the efforts imo.
- There seems to be no existing solid, lightweight alternative to SocketServer.
- The Python 3.5 socketserver module looks good enough on performance and
feature set. We don't gain much re-inventing it and we may fall behind stdlib
in the long run.
It seems reasonable although suboptimal to do a backport now. This backport and
its following changes will maintain API compatibility with Python's standard
socketserver. Eventually when we only need to support Python >= 3.5, we can
easily remove this file and just use Python's.
Note that the file is left unmodified intensionally to make it easier to audit
or trace changes. It will break some test-check-*.t. The next patch will fix
these test.
diff --git a/mercurial/socketserver.py b/mercurial/socketserver.py
new file mode 100644
--- /dev/null
+++ b/mercurial/socketserver.py
@@ -0,0 +1,763 @@
+"""Generic socket server classes.
+
+This module tries to capture the various aspects of defining a server:
+
+For socket-based servers:
+
+- address family:
+ - AF_INET{,6}: IP (Internet Protocol) sockets (default)
+ - AF_UNIX: Unix domain sockets
+ - others, e.g. AF_DECNET are conceivable (see <socket.h>
+- socket type:
+ - SOCK_STREAM (reliable stream, e.g. TCP)
+ - SOCK_DGRAM (datagrams, e.g. UDP)
+
+For request-based servers (including socket-based):
+
+- client address verification before further looking at the request
+ (This is actually a hook for any processing that needs to look
+ at the request before anything else, e.g. logging)
+- how to handle multiple requests:
+ - synchronous (one request is handled at a time)
+ - forking (each request is handled by a new process)
+ - threading (each request is handled by a new thread)
+
+The classes in this module favor the server type that is simplest to
+write: a synchronous TCP/IP server. This is bad class design, but
+save some typing. (There's also the issue that a deep class hierarchy
+slows down method lookups.)
+
+There are five classes in an inheritance diagram, four of which represent
+synchronous servers of four types:
+
+ +------------+
+ | BaseServer |
+ +------------+
+ |
+ v
+ +-----------+ +------------------+
+ | TCPServer |------->| UnixStreamServer |
+ +-----------+ +------------------+
+ |
+ v
+ +-----------+ +--------------------+
+ | UDPServer |------->| UnixDatagramServer |
+ +-----------+ +--------------------+
+
+Note that UnixDatagramServer derives from UDPServer, not from
+UnixStreamServer -- the only difference between an IP and a Unix
+stream server is the address family, which is simply repeated in both
+unix server classes.
+
+Forking and threading versions of each type of server can be created
+using the ForkingMixIn and ThreadingMixIn mix-in classes. For
+instance, a threading UDP server class is created as follows:
+
+ class ThreadingUDPServer(ThreadingMixIn, UDPServer): pass
+
+The Mix-in class must come first, since it overrides a method defined
+in UDPServer! Setting the various member variables also changes
+the behavior of the underlying server mechanism.
+
+To implement a service, you must derive a class from
+BaseRequestHandler and redefine its handle() method. You can then run
+various versions of the service by combining one of the server classes
+with your request handler class.
+
+The request handler class must be different for datagram or stream
+services. This can be hidden by using the request handler
+subclasses StreamRequestHandler or DatagramRequestHandler.
+
+Of course, you still have to use your head!
+
+For instance, it makes no sense to use a forking server if the service
+contains state in memory that can be modified by requests (since the
+modifications in the child process would never reach the initial state
+kept in the parent process and passed to each child). In this case,
+you can use a threading server, but you will probably have to use
+locks to avoid two requests that come in nearly simultaneous to apply
+conflicting changes to the server state.
+
+On the other hand, if you are building e.g. an HTTP server, where all
+data is stored externally (e.g. in the file system), a synchronous
+class will essentially render the service "deaf" while one request is
+being handled -- which may be for a very long time if a client is slow
+to read all the data it has requested. Here a threading or forking
+server is appropriate.
+
+In some cases, it may be appropriate to process part of a request
+synchronously, but to finish processing in a forked child depending on
+the request data. This can be implemented by using a synchronous
+server and doing an explicit fork in the request handler class
+handle() method.
+
+Another approach to handling multiple simultaneous requests in an
+environment that supports neither threads nor fork (or where these are
+too expensive or inappropriate for the service) is to maintain an
+explicit table of partially finished requests and to use a selector to
+decide which request to work on next (or whether to handle a new
+incoming request). This is particularly important for stream services
+where each client can potentially be connected for a long time (if
+threads or subprocesses cannot be used).
+
+Future work:
+- Standard classes for Sun RPC (which uses either UDP or TCP)
+- Standard mix-in classes to implement various authentication
+ and encryption schemes
+
+XXX Open problems:
+- What to do with out-of-band data?
+
+BaseServer:
+- split generic "request" functionality out into BaseServer class.
+ Copyright (C) 2000 Luke Kenneth Casson Leighton <lkcl at samba.org>
+
+ example: read entries from a SQL database (requires overriding
+ get_request() to return a table entry from the database).
+ entry is processed by a RequestHandlerClass.
+
+"""
+
+# Author of the BaseServer patch: Luke Kenneth Casson Leighton
+
+# XXX Warning!
+# There is a test suite for this module, but it cannot be run by the
+# standard regression test.
+# To run it manually, run Lib/test/test_socketserver.py.
+
+__version__ = "0.4"
+
+
+import socket
+import selectors
+import os
+import errno
+try:
+ import threading
+except ImportError:
+ import dummy_threading as threading
+from time import monotonic as time
+
+__all__ = ["BaseServer", "TCPServer", "UDPServer", "ForkingUDPServer",
+ "ForkingTCPServer", "ThreadingUDPServer", "ThreadingTCPServer",
+ "BaseRequestHandler", "StreamRequestHandler",
+ "DatagramRequestHandler", "ThreadingMixIn", "ForkingMixIn"]
+if hasattr(socket, "AF_UNIX"):
+ __all__.extend(["UnixStreamServer","UnixDatagramServer",
+ "ThreadingUnixStreamServer",
+ "ThreadingUnixDatagramServer"])
+
+# poll/select have the advantage of not requiring any extra file descriptor,
+# contrarily to epoll/kqueue (also, they require a single syscall).
+if hasattr(selectors, 'PollSelector'):
+ _ServerSelector = selectors.PollSelector
+else:
+ _ServerSelector = selectors.SelectSelector
+
+
+class BaseServer:
+
+ """Base class for server classes.
+
+ Methods for the caller:
+
+ - __init__(server_address, RequestHandlerClass)
+ - serve_forever(poll_interval=0.5)
+ - shutdown()
+ - handle_request() # if you do not use serve_forever()
+ - fileno() -> int # for selector
+
+ Methods that may be overridden:
+
+ - server_bind()
+ - server_activate()
+ - get_request() -> request, client_address
+ - handle_timeout()
+ - verify_request(request, client_address)
+ - server_close()
+ - process_request(request, client_address)
+ - shutdown_request(request)
+ - close_request(request)
+ - service_actions()
+ - handle_error()
+
+ Methods for derived classes:
+
+ - finish_request(request, client_address)
+
+ Class variables that may be overridden by derived classes or
+ instances:
+
+ - timeout
+ - address_family
+ - socket_type
+ - allow_reuse_address
+
+ Instance variables:
+
+ - RequestHandlerClass
+ - socket
+
+ """
+
+ timeout = None
+
+ def __init__(self, server_address, RequestHandlerClass):
+ """Constructor. May be extended, do not override."""
+ self.server_address = server_address
+ self.RequestHandlerClass = RequestHandlerClass
+ self.__is_shut_down = threading.Event()
+ self.__shutdown_request = False
+
+ def server_activate(self):
+ """Called by constructor to activate the server.
+
+ May be overridden.
+
+ """
+ pass
+
+ def serve_forever(self, poll_interval=0.5):
+ """Handle one request at a time until shutdown.
+
+ Polls for shutdown every poll_interval seconds. Ignores
+ self.timeout. If you need to do periodic tasks, do them in
+ another thread.
+ """
+ self.__is_shut_down.clear()
+ try:
+ # XXX: Consider using another file descriptor or connecting to the
+ # socket to wake this up instead of polling. Polling reduces our
+ # responsiveness to a shutdown request and wastes cpu at all other
+ # times.
+ with _ServerSelector() as selector:
+ selector.register(self, selectors.EVENT_READ)
+
+ while not self.__shutdown_request:
+ ready = selector.select(poll_interval)
+ if ready:
+ self._handle_request_noblock()
+
+ self.service_actions()
+ finally:
+ self.__shutdown_request = False
+ self.__is_shut_down.set()
+
+ def shutdown(self):
+ """Stops the serve_forever loop.
+
+ Blocks until the loop has finished. This must be called while
+ serve_forever() is running in another thread, or it will
+ deadlock.
+ """
+ self.__shutdown_request = True
+ self.__is_shut_down.wait()
+
+ def service_actions(self):
+ """Called by the serve_forever() loop.
+
+ May be overridden by a subclass / Mixin to implement any code that
+ needs to be run during the loop.
+ """
+ pass
+
+ # The distinction between handling, getting, processing and finishing a
+ # request is fairly arbitrary. Remember:
+ #
+ # - handle_request() is the top-level call. It calls selector.select(),
+ # get_request(), verify_request() and process_request()
+ # - get_request() is different for stream or datagram sockets
+ # - process_request() is the place that may fork a new process or create a
+ # new thread to finish the request
+ # - finish_request() instantiates the request handler class; this
+ # constructor will handle the request all by itself
+
+ def handle_request(self):
+ """Handle one request, possibly blocking.
+
+ Respects self.timeout.
+ """
+ # Support people who used socket.settimeout() to escape
+ # handle_request before self.timeout was available.
+ timeout = self.socket.gettimeout()
+ if timeout is None:
+ timeout = self.timeout
+ elif self.timeout is not None:
+ timeout = min(timeout, self.timeout)
+ if timeout is not None:
+ deadline = time() + timeout
+
+ # Wait until a request arrives or the timeout expires - the loop is
+ # necessary to accommodate early wakeups due to EINTR.
+ with _ServerSelector() as selector:
+ selector.register(self, selectors.EVENT_READ)
+
+ while True:
+ ready = selector.select(timeout)
+ if ready:
+ return self._handle_request_noblock()
+ else:
+ if timeout is not None:
+ timeout = deadline - time()
+ if timeout < 0:
+ return self.handle_timeout()
+
+ def _handle_request_noblock(self):
+ """Handle one request, without blocking.
+
+ I assume that selector.select() has returned that the socket is
+ readable before this function was called, so there should be no risk of
+ blocking in get_request().
+ """
+ try:
+ request, client_address = self.get_request()
+ except OSError:
+ return
+ if self.verify_request(request, client_address):
+ try:
+ self.process_request(request, client_address)
+ except:
+ self.handle_error(request, client_address)
+ self.shutdown_request(request)
+
+ def handle_timeout(self):
+ """Called if no new request arrives within self.timeout.
+
+ Overridden by ForkingMixIn.
+ """
+ pass
+
+ def verify_request(self, request, client_address):
+ """Verify the request. May be overridden.
+
+ Return True if we should proceed with this request.
+
+ """
+ return True
+
+ def process_request(self, request, client_address):
+ """Call finish_request.
+
+ Overridden by ForkingMixIn and ThreadingMixIn.
+
+ """
+ self.finish_request(request, client_address)
+ self.shutdown_request(request)
+
+ def server_close(self):
+ """Called to clean-up the server.
+
+ May be overridden.
+
+ """
+ pass
+
+ def finish_request(self, request, client_address):
+ """Finish one request by instantiating RequestHandlerClass."""
+ self.RequestHandlerClass(request, client_address, self)
+
+ def shutdown_request(self, request):
+ """Called to shutdown and close an individual request."""
+ self.close_request(request)
+
+ def close_request(self, request):
+ """Called to clean up an individual request."""
+ pass
+
+ def handle_error(self, request, client_address):
+ """Handle an error gracefully. May be overridden.
+
+ The default is to print a traceback and continue.
+
+ """
+ print('-'*40)
+ print('Exception happened during processing of request from', end=' ')
+ print(client_address)
+ import traceback
+ traceback.print_exc() # XXX But this goes to stderr!
+ print('-'*40)
+
+
+class TCPServer(BaseServer):
+
+ """Base class for various socket-based server classes.
+
+ Defaults to synchronous IP stream (i.e., TCP).
+
+ Methods for the caller:
+
+ - __init__(server_address, RequestHandlerClass, bind_and_activate=True)
+ - serve_forever(poll_interval=0.5)
+ - shutdown()
+ - handle_request() # if you don't use serve_forever()
+ - fileno() -> int # for selector
+
+ Methods that may be overridden:
+
+ - server_bind()
+ - server_activate()
+ - get_request() -> request, client_address
+ - handle_timeout()
+ - verify_request(request, client_address)
+ - process_request(request, client_address)
+ - shutdown_request(request)
+ - close_request(request)
+ - handle_error()
+
+ Methods for derived classes:
+
+ - finish_request(request, client_address)
+
+ Class variables that may be overridden by derived classes or
+ instances:
+
+ - timeout
+ - address_family
+ - socket_type
+ - request_queue_size (only for stream sockets)
+ - allow_reuse_address
+
+ Instance variables:
+
+ - server_address
+ - RequestHandlerClass
+ - socket
+
+ """
+
+ address_family = socket.AF_INET
+
+ socket_type = socket.SOCK_STREAM
+
+ request_queue_size = 5
+
+ allow_reuse_address = False
+
+ def __init__(self, server_address, RequestHandlerClass, bind_and_activate=True):
+ """Constructor. May be extended, do not override."""
+ BaseServer.__init__(self, server_address, RequestHandlerClass)
+ self.socket = socket.socket(self.address_family,
+ self.socket_type)
+ if bind_and_activate:
+ try:
+ self.server_bind()
+ self.server_activate()
+ except:
+ self.server_close()
+ raise
+
+ def server_bind(self):
+ """Called by constructor to bind the socket.
+
+ May be overridden.
+
+ """
+ if self.allow_reuse_address:
+ self.socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+ self.socket.bind(self.server_address)
+ self.server_address = self.socket.getsockname()
+
+ def server_activate(self):
+ """Called by constructor to activate the server.
+
+ May be overridden.
+
+ """
+ self.socket.listen(self.request_queue_size)
+
+ def server_close(self):
+ """Called to clean-up the server.
+
+ May be overridden.
+
+ """
+ self.socket.close()
+
+ def fileno(self):
+ """Return socket file number.
+
+ Interface required by selector.
+
+ """
+ return self.socket.fileno()
+
+ def get_request(self):
+ """Get the request and client address from the socket.
+
+ May be overridden.
+
+ """
+ return self.socket.accept()
+
+ def shutdown_request(self, request):
+ """Called to shutdown and close an individual request."""
+ try:
+ #explicitly shutdown. socket.close() merely releases
+ #the socket and waits for GC to perform the actual close.
+ request.shutdown(socket.SHUT_WR)
+ except OSError:
+ pass #some platforms may raise ENOTCONN here
+ self.close_request(request)
+
+ def close_request(self, request):
+ """Called to clean up an individual request."""
+ request.close()
+
+
+class UDPServer(TCPServer):
+
+ """UDP server class."""
+
+ allow_reuse_address = False
+
+ socket_type = socket.SOCK_DGRAM
+
+ max_packet_size = 8192
+
+ def get_request(self):
+ data, client_addr = self.socket.recvfrom(self.max_packet_size)
+ return (data, self.socket), client_addr
+
+ def server_activate(self):
+ # No need to call listen() for UDP.
+ pass
+
+ def shutdown_request(self, request):
+ # No need to shutdown anything.
+ self.close_request(request)
+
+ def close_request(self, request):
+ # No need to close anything.
+ pass
+
+class ForkingMixIn:
+
+ """Mix-in class to handle each request in a new process."""
+
+ timeout = 300
+ active_children = None
+ max_children = 40
+
+ def collect_children(self):
+ """Internal routine to wait for children that have exited."""
+ if self.active_children is None:
+ return
+
+ # If we're above the max number of children, wait and reap them until
+ # we go back below threshold. Note that we use waitpid(-1) below to be
+ # able to collect children in size(<defunct children>) syscalls instead
+ # of size(<children>): the downside is that this might reap children
+ # which we didn't spawn, which is why we only resort to this when we're
+ # above max_children.
+ while len(self.active_children) >= self.max_children:
+ try:
+ pid, _ = os.waitpid(-1, 0)
+ self.active_children.discard(pid)
+ except ChildProcessError:
+ # we don't have any children, we're done
+ self.active_children.clear()
+ except OSError:
+ break
+
+ # Now reap all defunct children.
+ for pid in self.active_children.copy():
+ try:
+ pid, _ = os.waitpid(pid, os.WNOHANG)
+ # if the child hasn't exited yet, pid will be 0 and ignored by
+ # discard() below
+ self.active_children.discard(pid)
+ except ChildProcessError:
+ # someone else reaped it
+ self.active_children.discard(pid)
+ except OSError:
+ pass
+
+ def handle_timeout(self):
+ """Wait for zombies after self.timeout seconds of inactivity.
+
+ May be extended, do not override.
+ """
+ self.collect_children()
+
+ def service_actions(self):
+ """Collect the zombie child processes regularly in the ForkingMixIn.
+
+ service_actions is called in the BaseServer's serve_forver loop.
+ """
+ self.collect_children()
+
+ def process_request(self, request, client_address):
+ """Fork a new subprocess to process the request."""
+ pid = os.fork()
+ if pid:
+ # Parent process
+ if self.active_children is None:
+ self.active_children = set()
+ self.active_children.add(pid)
+ self.close_request(request)
+ return
+ else:
+ # Child process.
+ # This must never return, hence os._exit()!
+ try:
+ self.finish_request(request, client_address)
+ self.shutdown_request(request)
+ os._exit(0)
+ except:
+ try:
+ self.handle_error(request, client_address)
+ self.shutdown_request(request)
+ finally:
+ os._exit(1)
+
+
+class ThreadingMixIn:
+ """Mix-in class to handle each request in a new thread."""
+
+ # Decides how threads will act upon termination of the
+ # main process
+ daemon_threads = False
+
+ def process_request_thread(self, request, client_address):
+ """Same as in BaseServer but as a thread.
+
+ In addition, exception handling is done here.
+
+ """
+ try:
+ self.finish_request(request, client_address)
+ self.shutdown_request(request)
+ except:
+ self.handle_error(request, client_address)
+ self.shutdown_request(request)
+
+ def process_request(self, request, client_address):
+ """Start a new thread to process the request."""
+ t = threading.Thread(target = self.process_request_thread,
+ args = (request, client_address))
+ t.daemon = self.daemon_threads
+ t.start()
+
+
+class ForkingUDPServer(ForkingMixIn, UDPServer): pass
+class ForkingTCPServer(ForkingMixIn, TCPServer): pass
+
+class ThreadingUDPServer(ThreadingMixIn, UDPServer): pass
+class ThreadingTCPServer(ThreadingMixIn, TCPServer): pass
+
+if hasattr(socket, 'AF_UNIX'):
+
+ class UnixStreamServer(TCPServer):
+ address_family = socket.AF_UNIX
+
+ class UnixDatagramServer(UDPServer):
+ address_family = socket.AF_UNIX
+
+ class ThreadingUnixStreamServer(ThreadingMixIn, UnixStreamServer): pass
+
+ class ThreadingUnixDatagramServer(ThreadingMixIn, UnixDatagramServer): pass
+
+class BaseRequestHandler:
+
+ """Base class for request handler classes.
+
+ This class is instantiated for each request to be handled. The
+ constructor sets the instance variables request, client_address
+ and server, and then calls the handle() method. To implement a
+ specific service, all you need to do is to derive a class which
+ defines a handle() method.
+
+ The handle() method can find the request as self.request, the
+ client address as self.client_address, and the server (in case it
+ needs access to per-server information) as self.server. Since a
+ separate instance is created for each request, the handle() method
+ can define arbitrary other instance variariables.
+
+ """
+
+ def __init__(self, request, client_address, server):
+ self.request = request
+ self.client_address = client_address
+ self.server = server
+ self.setup()
+ try:
+ self.handle()
+ finally:
+ self.finish()
+
+ def setup(self):
+ pass
+
+ def handle(self):
+ pass
+
+ def finish(self):
+ pass
+
+
+# The following two classes make it possible to use the same service
+# class for stream or datagram servers.
+# Each class sets up these instance variables:
+# - rfile: a file object from which receives the request is read
+# - wfile: a file object to which the reply is written
+# When the handle() method returns, wfile is flushed properly
+
+
+class StreamRequestHandler(BaseRequestHandler):
+
+ """Define self.rfile and self.wfile for stream sockets."""
+
+ # Default buffer sizes for rfile, wfile.
+ # We default rfile to buffered because otherwise it could be
+ # really slow for large data (a getc() call per byte); we make
+ # wfile unbuffered because (a) often after a write() we want to
+ # read and we need to flush the line; (b) big writes to unbuffered
+ # files are typically optimized by stdio even when big reads
+ # aren't.
+ rbufsize = -1
+ wbufsize = 0
+
+ # A timeout to apply to the request socket, if not None.
+ timeout = None
+
+ # Disable nagle algorithm for this socket, if True.
+ # Use only when wbufsize != 0, to avoid small packets.
+ disable_nagle_algorithm = False
+
+ def setup(self):
+ self.connection = self.request
+ if self.timeout is not None:
+ self.connection.settimeout(self.timeout)
+ if self.disable_nagle_algorithm:
+ self.connection.setsockopt(socket.IPPROTO_TCP,
+ socket.TCP_NODELAY, True)
+ self.rfile = self.connection.makefile('rb', self.rbufsize)
+ self.wfile = self.connection.makefile('wb', self.wbufsize)
+
+ def finish(self):
+ if not self.wfile.closed:
+ try:
+ self.wfile.flush()
+ except socket.error:
+ # An final socket error may have occurred here, such as
+ # the local error ECONNABORTED.
+ pass
+ self.wfile.close()
+ self.rfile.close()
+
+
+class DatagramRequestHandler(BaseRequestHandler):
+
+ # XXX Regrettably, I cannot get this working on Linux;
+ # s.recvfrom() doesn't return a meaningful client address.
+
+ """Define self.rfile and self.wfile for datagram sockets."""
+
+ def setup(self):
+ from io import BytesIO
+ self.packet, self.socket = self.request
+ self.rfile = BytesIO(self.packet)
+ self.wfile = BytesIO()
+
+ def finish(self):
+ self.socket.sendto(self.wfile.getvalue(), self.client_address)
More information about the Mercurial-devel
mailing list