"""Generic socket server classes.
This module tries to capture the various aspects of defining a server:
For socket-based servers:
- AF_INET{,6}: IP (Internet Protocol) sockets (default)
- AF_UNIX: Unix domain sockets
- others, e.g. AF_DECNET are conceivable (see <socket.h>
- 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:
+-----------+ +------------------+
| TCPServer |------->| UnixStreamServer |
+-----------+ +------------------+
+-----------+ +--------------------+
| 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
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
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
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 select() 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).
- Standard classes for Sun RPC (which uses either UDP or TCP)
- Standard mix-in classes to implement various authentication
- Standard framework for select-based multiplexing
- What to do with out-of-band data?
- split generic "request" functionality out into BaseServer class.
Copyright (C) 2000 Luke Kenneth Casson Leighton <lkcl@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
import dummy_threading as threading
__all__ = ["TCPServer","UDPServer","ForkingUDPServer","ForkingTCPServer",
"ThreadingUDPServer","ThreadingTCPServer","BaseRequestHandler",
"StreamRequestHandler","DatagramRequestHandler",
"ThreadingMixIn", "ForkingMixIn"]
if hasattr(socket, "AF_UNIX"):
__all__.extend(["UnixStreamServer","UnixDatagramServer",
"ThreadingUnixStreamServer",
"ThreadingUnixDatagramServer"])
def _eintr_retry(func, *args):
"""restart a system call interrupted by EINTR"""
except (OSError, select.error) as e:
if e.args[0] != errno.EINTR:
"""Base class for server classes.
- __init__(server_address, RequestHandlerClass)
- serve_forever(poll_interval=0.5)
- handle_request() # if you do not use serve_forever()
- fileno() -> int # for select()
Methods that may be overridden:
- get_request() -> request, client_address
- verify_request(request, client_address)
- process_request(request, client_address)
- shutdown_request(request)
Methods for derived classes:
- finish_request(request, client_address)
Class variables that may be overridden by derived classes or
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.
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
self.__is_shut_down.clear()
while not self.__shutdown_request:
# 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.
r, w, e = _eintr_retry(select.select, [self], [], [],
self._handle_request_noblock()
self.__shutdown_request = False
self.__is_shut_down.set()
"""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
self.__shutdown_request = True
self.__is_shut_down.wait()
# The distinction between handling, getting, processing and
# finishing a request is fairly arbitrary. Remember:
# - handle_request() is the top-level call. It calls
# 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.
# Support people who used socket.settimeout() to escape
# handle_request before self.timeout was available.
timeout = self.socket.gettimeout()
elif self.timeout is not None:
timeout = min(timeout, self.timeout)
fd_sets = _eintr_retry(select.select, [self], [], [], timeout)
self._handle_request_noblock()
def _handle_request_noblock(self):
"""Handle one request, without blocking.
I assume that select.select has returned that the socket is
readable before this function was called, so there should be
no risk of blocking in get_request().
request, client_address = self.get_request()
if self.verify_request(request, client_address):
self.process_request(request, client_address)
self.handle_error(request, client_address)
self.shutdown_request(request)
self.shutdown_request(request)
def handle_timeout(self):
"""Called if no new request arrives within self.timeout.
Overridden by ForkingMixIn.
def verify_request(self, request, client_address):
"""Verify the request. May be overridden.
Return True if we should proceed with this request.
def process_request(self, request, client_address):
Overridden by ForkingMixIn and ThreadingMixIn.
self.finish_request(request, client_address)
self.shutdown_request(request)
"""Called to clean-up the server.
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."""
def handle_error(self, request, client_address):
"""Handle an error gracefully. May be overridden.
The default is to print a traceback and continue.
print 'Exception happened during processing of request from',
traceback.print_exc() # XXX But this goes to stderr!
class TCPServer(BaseServer):
"""Base class for various socket-based server classes.
Defaults to synchronous IP stream (i.e., TCP).
- __init__(server_address, RequestHandlerClass, bind_and_activate=True)
- serve_forever(poll_interval=0.5)
- handle_request() # if you don't use serve_forever()
- fileno() -> int # for select()
Methods that may be overridden:
- get_request() -> request, client_address
- verify_request(request, client_address)
- process_request(request, client_address)
- shutdown_request(request)
Methods for derived classes:
- finish_request(request, client_address)
Class variables that may be overridden by derived classes or
- request_queue_size (only for stream sockets)
address_family = socket.AF_INET
socket_type = socket.SOCK_STREAM
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,
"""Called by constructor to bind the socket.
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.
self.socket.listen(self.request_queue_size)
"""Called to clean-up the server.
"""Return socket file number.
Interface required by select().
return self.socket.fileno()
"""Get the request and client address from the socket.
return self.socket.accept()
def shutdown_request(self, request):
"""Called to shutdown and close an individual request."""
#explicitly shutdown. socket.close() merely releases
#the socket and waits for GC to perform the actual close.
request.shutdown(socket.SHUT_WR)
pass #some platforms may raise ENOTCONN here
self.close_request(request)
def close_request(self, request):
"""Called to clean up an individual request."""
class UDPServer(TCPServer):
allow_reuse_address = False
socket_type = socket.SOCK_DGRAM
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.
def shutdown_request(self, request):
# No need to shutdown anything.
It is recommended that you Edit text format, this type of Fix handles quite a lot in one request
