Can pyzmq publishers be operated from class instances? - python

I have following code for a Publisher, which instantiates a few class instances and publishes some messages.
However, I don't receive anything at Subscriber side.
Publisher
import zmq
import time
from multiprocessing import Process
class SendData:
def __init__(self, msg, port):
self.msg = msg
self.port = port
ctx = zmq.Context()
self.sock = ctx.socket(zmq.PUB)
self.sock.bind('tcp://127.0.0.1:'+str(self.port))
time.sleep(1)
def sender(self):
self.sock.send_json(self.msg)
def main():
for device, port in zip(['2.2.2.2', '5.5.5.5'],[5001, 5002]):
msg = {device:'Some random message'}
instance = SendData(device, port)
Process(target=instance.sender).start()
if __name__ == "__main__":
main()
Subscriber
import zmq
ctx = zmq.Context()
recv_sock1 = ctx.socket(zmq.SUB)
recv_sock1.connect('tcp://127.0.0.1:5001')
recv_sock1.setsockopt(zmq.SUBSCRIBE, '')
recv_sock2 = ctx.socket(zmq.SUB)
recv_sock2.connect('tcp://127.0.0.1:5002')
recv_sock2.setsockopt(zmq.SUBSCRIBE, '')
while True:
if recv_sock1.poll(10):
msg = recv_sock1.recv_json()
print msg
if recv_sock2.poll(10):
msg = recv_sock2.recv_json()
print msg
I had subscribers started before publishers could publish anything. Also, I can see TCP connections are in Established so connections are made.
pyzmq version 16.0.0
python version 2.7
Q1: Are 0mq publishers supported from class instances?
Q2: What am I missing?

As was said before, trying to share the ZeroMQ context between processes is the problem here and the solution by user3666197 will work.
However, I would suggest subclassing multiprocessing.Process in this case. That way, it is much clearer what part of the code is executed in which process. It also makes your code more readable and reusable.
The following code creates one sender process per device. The sender processes can be reused during the runtime of your program to send more data:
import multiprocessing
import queue
import zmq
import time
class Sender(multiprocessing.Process):
def __init__(self, port):
super(Sender, self).__init__()
self._port = port
self._messages = multiprocessing.Queue()
self._do_stop = multiprocesing.Event()
def run(self):
"""
This code is executed in a new process.
"""
ctx = zmq.Context()
sock = ctx.socket(zmq.PUB)
sock.bind("tcp://127.0.0.1:" + str(self._port))
while not self._do_stop.is_set():
try:
msg = self._message.get_nowait()
sock.send_json(msg)
except queue.Empty:
time.sleep(0.01)
def stop(self):
self._do_stop.set()
def send_message(self, msg):
self._messages.put(msg)
def main():
data = zip(['2.2.2.2', '5.5.5.5'], [5001, 5002])
# create senders
senders = {device: Sender(port) for device, port in data}
# start senders
for device in senders:
senders[device].start()
# send messages
for device, port in zip(['2.2.2.2', '5.5.5.5'],[5001, 5002]):
msg = {device: 'Some random message'}
senders[device].send_message(msg)
# do more stuff here....
# ... e.g. send more messages
# ...
# once you are finished, stop the subprocesses
for device in senders:
senders[device].stop()
I hope this helps solving your problem.

A1: Yes, they are.
A2: Conflicts of scope-of-use v/s Zero-sharing, one of ZeroMQ maxims
Once your original Publisher code is being executed in main(), the class instantiation process creates ( i.e. inside the main()-process scope-of-use ), via the .__init__() constructor-method, it's own Context() -instance, which thus belongs ( incl. all of it's derived child-objects ( sockets et al ) ) to this main()-process.
Next, the call to the Process(...) initiates another few processes, that receive the class-instances ( the pity is that these have already created ZeroMQ non-shareable toys ) from the main()-scope-of-use.
Solution?
A possible dirty quick hack could be to defer the ZeroMQ Context() instantiation -- yes, just move it from .__init__() to some .aDeferredSETUP() that will be executed specifically under different scope-of-use inside each of the spinned-of Process()-process, different from main()-process and you ought be done, as Zero-sharing is safely obeyed.
class SendData:
def __init__(self, msg, port):
self.msg = msg
self.port = port
self.NotSETUP = True
self.ctx = None
self.sock = None
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ L8R
# ctx = zmq.Context()
# self.sock = ctx.socket( zmq.PUB )
# self.sock.bind( 'tcp://127.0.0.1:' + str( self.port ) )
# time.sleep( 1 )
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ L8R
def sender( self ):
if self.NotSETUP:
self.aDeferredSETUP()
self.sock.send_json( self.msg )
def aDeferredSETUP( self ): # create I/O-threads in Process(), not main()
self.ctx = zmq.Context()
self.sock = self.ctx.socket( zmq.PUB )
self.sock.bind( 'tcp://127.0.0.1:' + str( self.port ) )
time.sleep( 1 )

Related

Python: How do I lock an object until all threads have done a thing

I am writing a simple threaded server that will send a message to all clients. I have an object that is reset after posting the change message, however I am having a hard time figuring out how to reset that object only after all threads have posted the change message.
To add some context to the problem. I am building a multi user Tkinter python app which connects to a remote database to retrieve information and the application needs to know when data changes so that when a user updates data, all other running instances of the app will get the update. From what I understand, MySQL does not support asynchronous application updates. Instead of running a query every 5 seconds on the database to see if there is a change, I am putting this code server side so that it will send a message to a socket on the client that a change has occurred on the database.
The main loop is just a dummy that will simulate a change
Here is my code:
import socket, threading, time, select, os
class dbMonitor:
isDBAltered = False
def postChange(self):
self.isDBAltered = True
def __str__(self):
return str(self.isDBAltered)
class ThreadedServer(object):
def __init__(self, port,dbMon):
self.port = port
self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self.sock.setblocking(0)
self.sock.bind((socket.gethostname(), self.port))
self.dbMon = dbMon
def listen(self):
self.sock.listen(100)
read_list = [self.sock]
while True:
read,write,error = select.select(read_list,[],[],1)
for s in read:
if s is self.sock:
client, address = self.sock.accept()
client.settimeout(60)
threading.Thread(target = self.listenToClient, args = (client,address)).start()
def listenToClient(self, client, address):
read_list = [client]
size = 1024
while True:
response = b'Ack'
if self.dbMon.isDBAltered:
response = b'CHANGE'
try:
client.send(response)
except:
client.close()
return False
self.dbMon.isDBAltered = False
read,write,error = select.select(read_list,[],[],1)
for s in read:
if s is client:
try:
data = client.recv(size)
print(data)
if data:
client.send(response)
else:
raise error('Client disconnected')
except:
client.close()
return False
def mainLoop():
while True:
time.sleep(15)
print(dbMon)
dbMon.postChange()
dbMon = dbMonitor()
server = ThreadedServer(5005,dbMon)
threading.Thread(target = mainLoop, args=()).start()
threading.Thread(target = server.listen(), args=()).start()
How do I get self.dbMon.isDBAltered = False to execute only after all threads have executed:
response = b'CHANGE'
try:
client.send(response)
You're trying to synchronize something that's asynchronous... This is massively more complicated than it should be. Your dbmon is only storing a boolean flag... why not just asynchronously modify the "database" instead? For example, if the "database" was a thread-safe buffer, you could just append to that buffer or modify that buffer without synchronizing each thread individually, pull the information written to that buffer and write it to the client socket they belong to in another event loop (this is pretty much what asyncore does)
That said, I have some (probably nonworking, but I hope you get the idea) reference modified code for you to go off of if you want to continue pursing this avenue.
Basically, dbmon will keep a mapping of thread ids to [creation time, modified flag]
Our predicate returns true iff all threads created before a certain threshold have ALL set the modified flag. We set the modified flag when we send the response in the data = client.recv(size) portion of your code. And then we wait on that condition in the server send. We keep notifying all waiting threads on each client receive so that when the condition is finally met, our waiting server threads will all unblock and send the subsequent response.
import socket, threading, time, select, os
import collections
class dbMonitor:
def __init__(self):
self.isDBAltered = {}
self.lock = threading.Lock()
def newThread(self, tid):
self.lock.acquire()
# time of creation, boolean whether that thread has sent response
self.isDBAltered[tid] = [time.time(), False]
self.lock.release()
def threadDone(self, tid):
self.lock.acquire()
self.isDBAltered.pop(tid, None)
self.lock.release()
def altered(self, tid):
self.lock.acquire()
self.isDBAltered[tid][1] = True
self.lock.release()
def reset(self, tid):
self.lock.acquire()
self.isDBAltered(tid)[1] = False
self.lock.release()
def __str__(self):
return str(self.isDBAltered)
class ThreadedServer(object):
def __init__(self, port,dbMon):
self.port = port
self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self.sock.setblocking(0)
self.sock.bind((socket.gethostname(), self.port))
self.dbMon = dbMon
self.lock = threading.lock()
self.cv = threading.Condition()
self.thresh = 2000
def condition_pred(self):
# unblock if threads currently running for longer than self.thresh have all set their flags
return all([timecreate[1] if time.time() - timecreate[0] > self.thresh else True for tid,timecreate in self.dbMon.isDBAltered])
def listen(self):
self.sock.listen(100)
read_list = [self.sock]
while True:
read,write,error = select.select(read_list,[],[],1)
for s in read:
if s is self.sock:
self.lock.acquire()
client, address = self.sock.accept()
client.settimeout(60)
T = threading.Thread(target = self.listenToClient, args = (client,address)).start()
self.dbmon.newThread(T.ident)
self.lock.release()
def listenToClient(self, client, address):
read_list = [client]
size = 1024
while True:
response = b'Ack'
with self.cv:
self.cv.wait_for(self.condition_pred)
self.dbMon.reset(threading.get_ident())
response = b'CHANGE'
try:
client.send(response)
except:
client.close()
self.dbmon.threadDone(threading.get_ident())
return False
read,write,error = select.select(read_list,[],[],1)
for s in read:
if s is client:
with self.cv:
try:
data = client.recv(size)
print(data)
if data:
client.send(response)
self.dbMon.altered(threading.get_ident())
self.cv.notifyAll()
else:
raise error('Client disconnected')
except:
client.close()
self.dbmon.threadDone(threading.get_ident())
return False

Callback function does not see correct values in instance

I'm having a strange phenomena in Python with callback functions and handlers.
I use ZMQ to handle my communication and use a stream for the socket. I have the base class:
import multiprocessing
import zmq
from concurrent.futures import ThreadPoolExecutor
from zmq.eventloop import ioloop, zmqstream
from zmq.utils import jsonapi as json
# Types of messages
TYPE_A = 'type_a'
TYPE_B = 'type_b'
class ZmqProcess(multiprocessing.Process):
def __init__(self):
super(ZmqProcess, self).__init__()
self.context = None
self.loop = None
self.handle_stream = None
def setup(self):
self.context = zmq.Context()
self.loop = ioloop.IOLoop.instance()
def send(self, msg_type, msg, host, port):
sock = zmq.Context().socket(zmq.PAIR)
sock.connect('tcp://%s:%s' % (host, port))
sock.send_json([msg_type, msg])
def stream(self, sock_type, addr):
sock = self.context.socket(sock_type)
if isinstance(addr, str):
addr = addr.split(':')
host, port = addr if len(addr) == 2 else (addr[0], None)
if port:
sock.bind('tcp://%s:%s' % (host, port))
else:
port = sock.bind_to_random_port('tcp://%s' % host)
stream = zmqstream.ZMQStream(sock, self.loop)
return stream, int(port)
class MessageHandler(object):
def __init__(self, json_load=-1):
self._json_load = json_load
self.pool = ThreadPoolExecutor(max_workers=10)
def __call__(self, msg):
i = self._json_load
msg_type, data = json.loads(msg[i])
msg[i] = data
if msg_type.startswith('_'):
raise AttributeError('%s starts with an "_"' % msg_type)
getattr(self, msg_type)(*msg)
And I have a class that inherits from it:
import zmq
import zmq_base
class ZmqServerMeta(zmq_base.ZmqProcess):
def __init__(self, bind_addr, handlers):
super(ZmqServerMeta, self).__init__()
self.bind_addr = bind_addr
self.handlers = handlers
def setup(self):
super(ZmqServerMeta, self).setup()
self.handle_stream, _ = self.stream(zmq.PAIR, self.bind_addr)
self.handle_stream.on_recv(StreamHandler(self.handle_stream, self.stop,
self.handlers))
def run(self):
self.setup()
self.loop.start()
def stop(self):
self.loop.stop()
class StreamHandler(zmq_base.MessageHandler):
def __init__(self, handle_stream, stop, handlers):
super(StreamHandler, self).__init__()
self._handle_stream = handle_stream
self._stop = stop
self._handlers = handlers
def type_a(self, data):
if zmq_base.TYPE_A in self._handlers:
if self._handlers[zmq_base.TYPE_A]:
for handle in self._handlers[zmq_base.TYPE_A]:
self.pool.submit(handle, data)
else:
pass
else:
pass
def type_b(self, data):
if zmq_base.TYPE_B in self._handlers:
if self._handlers[zmq_base.TYPE_B]:
for handle in self._handlers[zmq_base.TYPE_B]:
self.pool.submit(handle, data)
else:
pass
else:
pass
def endit(self):
self._stop()
Additionally, I have a class that I want to use as storage. And here is where the trouble starts:
import threading
import zmq_server_meta as server
import zmq_base as base
class Storage:
def __init__(self):
self.list = []
self.list_lock = threading.RLock()
self.zmq_server = None
self.host = '127.0.0.1'
self.port = 5432
self.bind_addr = (self.host, self.port)
def setup(self):
handlers = {base.TYPE_A: [self. remove]}
self.zmq_server = server.ZmqServerMeta(handlers=handlers, bind_addr=self.bind_addr)
self.zmq_server.start()
def add(self, data):
with self.list_lock:
try:
self.list.append(data)
except:
print "Didn't work"
def remove(self, msg):
with self.list_lock:
try:
self.list.remove(msg)
except:
print "Didn't work"
The idea is that that class stores some global information that it receives.
It is all started in a file to test:
import sys
import time
import storage
import zmq_base as base
import zmq_server_meta as server
def printMsg(msg):
print msg
store = storage.Storage()
store.setup()
handlers = {base.TYPE_B: [printMsg]}
client = server.ZmqServerMeta(handlers=handlers, bind_addr=('127.0.0.1', 5431))
client.start()
message = "Test"
store.add(message)
client.send(base.TYPE_A, message, '127.0.0.1', 5432)
I simplified it to reduce clutter. Instead of just adding it, it is usually send and then a response comes back. The response, the client sending, should be processed by the correct callback, remove(), and it should remove something out of the list. The problem that occurs is, that the remove() function sees an empty list, although there should be an element in the list. If I check from the testing file, I can see the element after it was added, and if I call remove() from there, I see a non-empty list and can remove it. My question is, why does the callback sees an empty list and how can I make sure it does see the correct elements in the list?
Kind regards
Patrick
I believe the problem lays in the fact that the ZmqProcess class inherits from multiprocessing.Process. Multiprocessing does not allow to share objects among different processes, except by using a shared memory map using Value or Array ( as can be seen in the documentation: https://docs.python.org/3/library/multiprocessing.html#sharing-state-between-processes )
If you want to use your custom object, you can use a Server process / proxy object, which can be found in on the same page of the documentation.
So you can, for instance, define a manager in the init function of the Storage class like: self.manager = Manager() Afterwards you put self.list = self.manager.list(). This should do the trick.

too many file descriptors in select() python in windows

i am trying to receive about 1000 connections to my server but it cannot receive more than 512. What can i do to increase the amount of open connections? I am running windows 8.1
Not: I am very new to this stuff so, thanks for help
Here is my code;
import asyncore
import socket
import uuid
import time
import threading
class statistics(threading.Thread):
def __init__(self):
threading.Thread.__init__(self)
def run(self):
while True:
entry = raw_input("")
zaman = int(time.time())
cmd = receivedCmd
print "calculating.."
time.sleep(1)
if entry == 'istatistik':
print str(receivedCmd-cmd) + " command/second"
print "total received commands: " + str(receivedCmd)
entry = ""
class tcpClient:
def __init__(self):
self.clientid = uuid.uuid1(int(time.time()))
self.buffer = ""
self.buffer_size = 0
self.conn_time = time.time()
self.overflow = 0
#print str(self.clientid) + " assingned"
def recv_msg(self, msg):
global receivedCmd
self.buffer = msg
self.buffer_size = len(self.buffer)
receivedCmd = receivedCmd + 1
if self.buffer_size >= 1024:
self.overflow = 1
def __del__(self):
print str(self.clientid) + " has left."
class TCPHandler(asyncore.dispatcher_with_send):
global clist
def handle_read(self):
data = self.recv(1024)
if data:
if clist[self].overflow:
self.send("overflow")
self.handle_close()
else:
self.send(data)
clist[self].recv_msg(data)
def handle_close(self):
del clist[self]
self.close()
def handle_error(self):
del clist[self]
self.close()
class TCPServer(asyncore.dispatcher):
global clist
def __init__(self, host, port):
asyncore.dispatcher.__init__(self)
self.create_socket(socket.AF_INET, socket.SOCK_STREAM)
self.set_reuse_addr()
self.bind((host, port))
self.listen(5)
def handle_accept(self):
self.clist = clist
pair = self.accept()
if pair is None:
pass
else:
sock, addr = pair
#print 'Connection : %s' % repr(addr)
clist[TCPHandler(sock)] = tcpClient()
if __name__ == '__main__':
clist = {}
receivedCmd = 0
server = TCPServer('', 5000)
server2 = TCPServer('',5001)
StaticsThread = statistics()
StaticsThread.start()
asyncore.loop()
Note: I still cannot receive more than 512 connections with the Twisted Framework, i don't know what to do. There have to be thousands of connected clients. Please help.
The asyncore module relies in the select OS function, which only supports a limited number of file descriptors.
As an alternative use a multi-threading server (I won't recommend this) or, better, the Twisted framework which is event-driven (highly recommended!).
Hope this helps!
Since Twisted's default reactor under Windows is also select-based then you should consider using the IOCP reactor instead.
from twisted.internet import iocpreactor
iocpreactor.install()
from twisted.internet import reactor
But also take into account that Twisted prefers Linux systems (where the default reactor is epoll-based) rather than Windows. Maybe switching to Linux is a better choice.

zeromq: why dealer sends one message and receives from one client?

I'm trying to understand my code's behavior.
I'm using zeromq to create a server that sends a "ping" and waits for "pong" responses.
What i'm seeing is that when I send a ping, only one client receives it.
when I run this code and send "ping" for the first time i receive:
pong: A
and when i run it again, i get
pong: B
why is that? I want to send one "ping" and receive two pongs.
here's the code:
from threading import Thread
import zmq
class zmqdealer(object):
def __init__(self, port):
context = zmq.Context()
self.sock = context.socket(zmq.DEALER)
#self.sock.setsockopt(zmq.RCVTIMEO, 1000)
self.sock.bind("tcp://*:%s" % port)
thread = Thread(target=lambda: self.poll())
thread.daemon = True
thread.start()
def poll(self):
while True:
reply = self.sock.recv()
if reply != "":
print(reply)
def ping(self):
self.sock.send_multipart(['', 'ping'])
class zmqrep(object):
def __init__(self, ident,host, port):
context = zmq.Context()
self.sock = context.socket(zmq.REP)
self.sock.connect("tcp://%s:%s" % (host, port))
self.ident = ident
thread = Thread(target=lambda: self.pong())
thread.daemon = True
thread.start()
def pong(self):
while True:
request = self.sock.recv()
if request == "ping":
msg = "pong: %s" % self.ident
self.sock.send(msg)
if __name__ == "__main__":
port = 11112
host = "localhost"
server = zmqdealer(port)
client1 = zmqrep('A',host,port)
client2 = zmqrep('B',host,port)
answer = raw_input('press <ENTER> to exit or type \'ping\' to get a pong\n')
while True:
if answer == "":
break
if answer == "ping":
server.ping()
answer = raw_input()
EDIT
I found a way to make this work. I really hope there is another way because i genuinely hate this one! so it looks like dealer sends to the clients in a round robin fashion. so to make my ping work i had to send it to all the clients. how? i subscribed to the monitor socket and added every connected client to a list. every time i ping, i ping to every client. look:
import threading
import zmq
from zmq.utils import monitor
def threadify(func, daemon=True):
thread = threading.Thread(target=func)
thread.daemon = daemon
thread.start()
class zmqdealer(object):
def __init__(self, port):
context = zmq.Context()
self.sock = context.socket(zmq.DEALER)
self.monitor_sock = self.sock.get_monitor_socket()
self.sock.bind("tcp://*:%s" % port)
self.connected_clients = {}
threadify(func=self.poll)
threadify(func=self.monitor)
def poll(self):
while True:
reply = self.sock.recv()
if reply != "":
print reply
def add_client(self, event):
endpoint = event['endpoint']
value = event['value']
if endpoint in self.connected_clients:
self.connected_clients[endpoint].append(value)
else:
self.connected_clients[endpoint] = [value]
def remove_client(self, event):
endpoint = event['endpoint']
value = event['value']
if endpoint in self.connected_clients \
and value in self.connected_clients[endpoint]:
self.connected_clients[endpoint].remove(value)
def monitor(self):
options = {zmq.EVENT_ACCEPTED: lambda e: self.add_client(e),
zmq.EVENT_DISCONNECTED: lambda e: self.remove_client(e)}
while True:
event = monitor.recv_monitor_message(self.monitor_sock)
event_type = event['event']
if event_type in options:
options[event_type](event)
event['event'] = event_types[event_type]
print event
def ping(self):
connected_clients_amount = sum([len(clients) for clients in self.connected_clients.values()])
for i in xrange(connected_clients_amount):
self.sock.send_multipart(['', 'ping'])
if connected_clients_amount <= 0:
print "there are no connected clients!"
class zmqrep(object):
def __init__(self, ident, host, port):
context = zmq.Context()
self.sock = context.socket(zmq.REP)
self.sock.connect("tcp://%s:%s" % (host, port))
self.identity = ident
self.stopped = threading.Event()
threadify(self.pong)
def pong(self):
while not self.stopped.isSet():
request = self.sock.recv()
if request == "ping":
msg = "pong: %s" % self.identity
self.sock.send(msg)
self.sock.close()
def stop(self):
self.stopped.set()
if __name__ == "__main__":
port = 11112
host = "localhost"
num = 5
server = zmqdealer(port)
clients = [zmqrep(i.__str__(), host, port) for i in xrange(num)]
answer = raw_input('press <ENTER> to exit or type \'ping\' to get a pong\n')
while True:
if answer == "":
break
if answer == "ping":
server.ping()
if answer == "kill":
if len(clients) > 0:
die = clients[0]
clients.remove(die)
die.stop()
else:
print "there are no connected clients!\n"
answer = raw_input()
Router/Dealer sockets are best used for distributing tasks. Say you have 10 tasks and 2 workers, you do not care who does what. Dealer/Router will distribute in a round robin fashion.
Maybe Pub/Sub or Push/Pull sockets would fit your usecase better? They are both broadcast sockets.
Here's an example of Push/Pull used in a similar fashion as what you're doing.
You often end up doing pairs of sockets, one to transmit and one other to receive results. You could for example do a PUSH with a ping message + random identifier, and ask clients to answer on PUB/SUB where you subscribe to this random identifier. This way you can match requests and responses.

Python multiprocessing RemoteManager under a multiprocessing.Process

I'm trying to start a data queue server under a managing process (so that it can later be turned into a service), and while the data queue server function works fine in the main process, it does not work in a process created using multiprocessing.Process.
The dataQueueServer and dataQueueClient code is based on the code from the multiprocessing module documentation here.
When run on its own, dataQueueServer works well. However, when run using a multiprocessing.Process's start() in mpquueue, it doesn't work (when tested with the client). I am using the dataQueueClient without changes to test both cases.
The code does reach the serve_forever in both cases, so I think the server is working, but something is blocking it from communicating back to the client in the mpqueue case.
I have placed the loop that runs the serve_forever() part under a thread, so that it can be stoppable.
Here is the code:
mpqueue # this is the "manager" process trying to spawn the server in a child process
import time
import multiprocessing
import threading
import dataQueueServer
class Printer():
def __init__(self):
self.lock = threading.Lock()
def tsprint(self, text):
with self.lock:
print text
class QueueServer(multiprocessing.Process):
def __init__(self, name = '', printer = None):
multiprocessing.Process.__init__(self)
self.name = name
self.printer = printer
self.ml = dataQueueServer.MainLoop(name = 'ml', printer = self.printer)
def run(self):
self.printer.tsprint(self.ml)
self.ml.start()
def stop(self):
self.ml.stop()
if __name__ == '__main__':
printer = Printer()
qs = QueueServer(name = 'QueueServer', printer = printer)
printer.tsprint(qs)
printer.tsprint('starting')
qs.start()
printer.tsprint('started.')
printer.tsprint('Press Ctrl-C to quit')
try:
while True:
time.sleep(60)
except KeyboardInterrupt:
printer.tsprint('\nTrying to exit cleanly...')
qs.stop()
printer.tsprint('stopped')
dataQueueServer
import time
import threading
from multiprocessing.managers import BaseManager
from multiprocessing import Queue
HOST = ''
PORT = 50010
AUTHKEY = 'authkey'
## Define some helper functions for use by the main process loop
class Printer():
def __init__(self):
self.lock = threading.Lock()
def tsprint(self, text):
with self.lock:
print text
class QueueManager(BaseManager):
pass
class MainLoop(threading.Thread):
"""A thread based loop manager, allowing termination signals to be sent
to the thread"""
def __init__(self, name = '', printer = None):
threading.Thread.__init__(self)
self._stopEvent = threading.Event()
self.daemon = True
self.name = name
if printer is None:
self.printer = Printer()
else:
self.printer = printer
## create the queue
self.queue = Queue()
## Add a function to the handler to return the queue to clients
self.QM = QueueManager
self.QM.register('get_queue', callable=lambda:self.queue)
self.queue_manager = self.QM(address=(HOST, PORT), authkey=AUTHKEY)
self.queue_server = self.queue_manager.get_server()
def __del__(self):
self.printer.tsprint( 'closing...')
def run(self):
self.printer.tsprint( '{}: started serving'.format(self.name))
self.queue_server.serve_forever()
def stop(self):
self.printer.tsprint ('{}: stopping'.format(self.name))
self._stopEvent.set()
def stopped(self):
return self._stopEvent.isSet()
def start():
printer = Printer()
ml = MainLoop(name = 'ml', printer = printer)
ml.start()
return ml
def stop(ml):
ml.stop()
if __name__ == '__main__':
ml = start()
raw_input("\nhit return to stop")
stop(ml)
And a client:
dataQueueClient
import datetime
from multiprocessing.managers import BaseManager
n = 0
N = 10**n
HOST = ''
PORT = 50010
AUTHKEY = 'authkey'
def now():
return datetime.datetime.now()
def gen(n, func, *args, **kwargs):
k = 0
while k < n:
yield func(*args, **kwargs)
k += 1
class QueueManager(BaseManager):
pass
QueueManager.register('get_queue')
m = QueueManager(address=(HOST, PORT), authkey=AUTHKEY)
m.connect()
queue = m.get_queue()
def load(msg, q):
return q.put(msg)
def get(q):
return q.get()
lgen = gen(N, load, msg = 'hello', q = queue)
t0 = now()
while True:
try:
lgen.next()
except StopIteration:
break
t1 = now()
print 'loaded %d items in ' % N, t1-t0
t0 = now()
while queue.qsize() > 0:
queue.get()
t1 = now()
print 'got %d items in ' % N, t1-t0
So it seems like the solution is simple enough: Don't use serve_forever(), and use manager.start() instead.
According to Eli Bendersky, the BaseManager (and it's extended version SyncManager) already spawns the server in a new process (and looking at the multiprocessing.managers code confirms this). The problem I have been experiencing stems from the form used in the example, in which the server is started under the main process.
I still don't understand why the current example doesn't work when run under a child process, but that's no longer an issue.
Here's the working (and much simplified from OP) code to manage multiple queue servers:
Server:
from multiprocessing import Queue
from multiprocessing.managers import SyncManager
HOST = ''
PORT0 = 5011
PORT1 = 5012
PORT2 = 5013
AUTHKEY = 'authkey'
name0 = 'qm0'
name1 = 'qm1'
name2 = 'qm2'
description = 'Queue Server'
def CreateQueueServer(HOST, PORT, AUTHKEY, name = None, description = None):
name = name
description = description
q = Queue()
class QueueManager(SyncManager):
pass
QueueManager.register('get_queue', callable = lambda: q)
QueueManager.register('get_name', callable = name)
QueueManager.register('get_description', callable = description)
manager = QueueManager(address = (HOST, PORT), authkey = AUTHKEY)
manager.start() # This actually starts the server
return manager
# Start three queue servers
qm0 = CreateQueueServer(HOST, PORT0, AUTHKEY, name0, description)
qm1 = CreateQueueServer(HOST, PORT1, AUTHKEY, name1, description)
qm2 = CreateQueueServer(HOST, PORT2, AUTHKEY, name2, description)
raw_input("return to end")
Client:
from multiprocessing.managers import SyncManager
HOST = ''
PORT0 = 5011
PORT1 = 5012
PORT2 = 5013
AUTHKEY = 'authkey'
def QueueServerClient(HOST, PORT, AUTHKEY):
class QueueManager(SyncManager):
pass
QueueManager.register('get_queue')
QueueManager.register('get_name')
QueueManager.register('get_description')
manager = QueueManager(address = (HOST, PORT), authkey = AUTHKEY)
manager.connect() # This starts the connected client
return manager
# create three connected managers
qc0 = QueueServerClient(HOST, PORT0, AUTHKEY)
qc1 = QueueServerClient(HOST, PORT1, AUTHKEY)
qc2 = QueueServerClient(HOST, PORT2, AUTHKEY)
# Get the queue objects from the clients
q0 = qc0.get_queue()
q1 = qc1.get_queue()
q2 = qc2.get_queue()
# put stuff in the queues
q0.put('some stuff')
q1.put('other stuff')
q2.put({1:123, 2:'abc'})
# check their sizes
print 'q0 size', q0.qsize()
print 'q1 size', q1.qsize()
print 'q2 size', q2.qsize()
# pull some stuff and print it
print q0.get()
print q1.get()
print q2.get()
Adding an additional server to share a dictionary with the information of the running queue servers so that consumers can easily tell what's available where is easy enough using that model. One thing to note, though, is that the shared dictionary requires slightly different syntax than a normal dictionary: dictionary[0] = something will not work. You need to use dictionary.update([(key, value), (otherkey, othervalue)]) and dictionary.get(key) syntax, which propagates across to all other clients connected to this dictionary..

Categories