Frage

I am using asyncore to implement a publish subscribe. I understand using twisted or ZMQ may be a better solution, however in this case it needs to be pure python. When waiting for connections the CPU usage is ~1%, as soon as a client connects the CPU usage jumps write up to 100%. It does not drop back down even after the client disconnects.

My server class:

class Host(asyncore.dispatcher):

    log = logging.getLogger('Host')

    def __init__(self, port=7655):
        asyncore.dispatcher.__init__(self)
        self.create_socket(socket.AF_INET, socket.SOCK_STREAM)
        self.set_reuse_addr()
        self.bind(('0.0.0.0', port,))
        self.listen(5)
        self.clients = []

    def handle_accept(self):
        socket, addr = self.accept()
        self.log.info("Aceepted client at {0}:{1}".format(addr[0], addr[1]))
        self.clients.append(RemoteClient(self, socket, addr))

    def broadcast(self, message):
        self.log.info("Broadcasting message: {0}".format(message))
        for client in self.clients:
            client.message(message)

And my handler:

class RemoteClient(asyncore.dispatcher):

    log = logging.getLogger('Host')

    def __init__(self, host, socket, address):
        asyncore.dispatcher.__init__(self, socket)
        self.host = host
        self.outbox = collections.deque()

    def message(self, message):
        self.outbox.append(message)

    def handle_write(self):
        if not self.outbox:
            return
        message = self.outbox.popleft()
        if len(message) > MAX_MESSAGE_LENGTH:
            raise ValueError('Message too long')
        self.send(message)

    def handle_close(self):
        self.host.clients.remove(self)
        self.log.info("Client removed from list")
        self.close()

    def handle_error(self):
        self.log.error("Socket error")

I have tried hunting for a solution but can't seem to work out what is going on. Any help appreciated!

War es hilfreich?

Lösung

Explanation

Your problem is that you don't override method asyncore.dispatcher.writeable, default implementation:

def writable(self):
    return True

Which causes asyncore.poll to run in while True loop (thus using 100% CPU):

def poll(timeout=0.0, map=None):
    if map is None:
        map = socket_map
    if map:
        r = []; w = []; e = []
        for fd, obj in map.items():
            is_r = obj.readable()
            is_w = obj.writable() # This is always true
            if is_r:
                r.append(fd)
            if is_w:
                w.append(fd)
            if is_r or is_w:
                e.append(fd) # This always happens
        if [] == r == w == e:
            time.sleep(timeout) # This never happens
            return

        try:
            # Here, having w (as function parameter) set to non-empty value
            # causes select to immediately return without waiting with w set
            # just to your client
            r, w, e = select.select(r, w, e, timeout) 

Solution

I think really clean solution would be re-implementing asyncore with some sort of threading.Event mechanism that would enforce waiting for writable object, but so far this worked for me:

# Add to RemoteClient
def writable(self):
    ''' It has point to call handle_write only when there's something in outbox
        Having this method always returning true will cause 100% CPU usage
    '''
    return bool(self.outbox)

# When you start loop
asyncore.loop(timeout=5.0) 

You also can see writable() overridden in example in official documentation.

I just prefer to be able to kill waiting sooner than after 30 seconds.

Lizenziert unter: CC-BY-SA mit Zuschreibung
Nicht verbunden mit StackOverflow
scroll top