Develop Reference

Why ZeroMQ fails to communicate when I use multiprocessing.Process to run?

please see the code below :
server.py
import zmq
import time
from multiprocessing import Process
class A:
def __init__(self):
ctx = zmq.Context(1)
sock = zmq.Socket(ctx, zmq.PUB)
sock.bind('ipc://test')
p = Process(target=A.run, args=(sock,))
p.start() # Process calls run, but the client can't receive messages
p.join() #
#A.run(sock) # this one is ok, messages get it to be received
#staticmethod
def run(sock):
while True:
sock.send('demo'.encode('utf-8'))
print('sent')
time.sleep(1)
if __name__ =='__main__':
a = A()
client.py
import zmq
ctx=zmq.Context(1)
sock = zmq.Socket(ctx, zmq.SUB)
sock.connect('ipc://test')
sock.setsockopt_string(zmq.SUBSCRIBE, '')
while True:
print(sock.recv())
In the constructor of server.py, if I call .run()-method directly, the client can receive the message, but when I use the multiprocessing.Process()-method, it fails. Can anyone explain on this and provide some advice?

Q : "Why ZeroMQ fails to communicate when I use multiprocessing.Process to run?"
Well, ZeroMQ does not fail to communicate, the problem is, how Python multiprocessing module "operates".
The module is designed so that some processing may escape from the python central GIL-lock (re-[SERIAL]-iser, that is used as a forever present [CONCURRENT]-situations' principal avoider).
This means that the call to the multiprocessing.Process makes one exact "mirror-copy" of the python interpreter state, "exported" into new O/S-spawned process (details depend on localhost O/S).
Given that, there is zero chance a "mirror"-ed replica could get access to resources already owned by the __main__ - here the .bind()-method already acquired ipc://test address, so "remote"-process will never get "permission" to touch this ZeroMQ AccessPoint, unless the code gets repaired & fully re-factored.
Q : "Can anyone explain on this and provide some advice?"
Sure. The best step to start is to fully understand Pythonic culture of monopolistic GIL-lock re-[SERIAL]-isation, where no two things ever happen in the same time, so even adding more threads does not speed-up the flow of the processing, as it all gets re-aligned by the central "monopolist" The GIL-lock.
Next, understanding a promise of a fully reflected copy of the python interpreter state, while it sounds promising, also has some obvious drawbacks - the new processes, being "mirror"-copies cannot introduce colliding cases on already owned resources. If they try to, a not working as expected cases are the milder of the problems in such principally ill-designed cases.
In your code, the first row in __main__ instantiates a = A(), where A's .__init__ method straight occupies the IPC-resource since .bind('ipc://test'). The later step, p = Process( target = A.run, args = ( sock, ) ) "mirror"-replicates the state of the python interpreter (an as-is copy) and the p.start() cannot but crash into disability to "own" the same resource as the __main__ already owns (yes, the ipc://test for a "mirror"-ed process instructed call to grab the same, non-free resource in .bind('ipc://test') ). This will never fly.
Last but not least, enjoy the Zen-of-Zero, the masterpiece of Martin SUSTRIK for distributed-computing, so well crafted for ultimately scalable, almost zero-latency, very comfortable, widely ported signalling & messaging framework.

Short answer: Start your subprocesses. Create your zmq.Context- and .Socket-instances from within your Producer.run()-classmethod within each subprocess. Use .bind()-method on the side on which your cardinality is 1, and .connect()-method on the side where your cardinality is >1 (in this case, the "server").
My approach would be structured something like...
# server.py :
import zmq
from multiprocessing import Process
class Producer (Process):
def init(self):
...
def run(self):
ctx = zmq.Context(1)
sock = zmq.Socket(ctx, zmq.PUB)
# Multiple producers, so connect instead of bind (consumer must bind)
sock.connect('ipc://test')
while True:
...
if __name__ == "__main__":
producer = Producer()
p = Process(target=producer.run)
p.start()
p.join()
# client.py :
import zmq
ctx = zmq.Context(1)
sock = zmq.Socket(ctx, zmq.SUB)
# Capture from multiple producers, so bind (producers must connect)
sock.bind('ipc://test')
sock.setsockopt_string(zmq.SUBSCRIBE, '')
while True:
print(sock.recv())

Separate computation from socket work in Python

I'm serializing column data and then sending it over a socket connection.
Something like:
import array, struct, socket
## Socket setup
s = socket.create_connection((ip, addr))
## Data container setup
ordered_col_list = ('col1', 'col2')
columns = dict.fromkeys(ordered_col_list)
for i in range(num_of_chunks):
## Binarize data
columns['col1'] = array.array('i', range(10000))
columns['col2'] = array.array('f', [float(num) for num in range(10000)])
.
.
.
## Send away
chunk = b''.join(columns[col_name] for col_name in ordered_col_list]
s.sendall(chunk)
s.recv(1000) #get confirmation
I wish to separate the computation from the sending, put them on separate threads or processes, so I can keep doing computations while data is sent away.
I've put the binarizing part as a generator function, then sent the generator to a separate thread, which then yielded binary chunks via a queue.
I collected the data from the main thread and sent it away. Something like:
import array, struct, socket
from time import sleep
try:
import thread
from Queue import Queue
except:
import _thread as thread
from queue import Queue
## Socket and queue setup
s = socket.create_connection((ip, addr))
chunk_queue = Queue()
def binarize(num_of_chunks):
''' Generator function that yields chunks of binary data. In reality it wouldn't be the same data'''
ordered_col_list = ('col1', 'col2')
columns = dict.fromkeys(ordered_col_list)
for i in range(num_of_chunks):
columns['col1'] = array.array('i', range(10000)).tostring()
columns['col2'] = array.array('f', [float(num) for num in range(10000)]).tostring()
.
.
yield b''.join((columns[col_name] for col_name in ordered_col_list))
def chunk_yielder(queue):
''' Generate binary chunks and put them on a queue. To be used from a thread '''
while True:
try:
data_gen = queue.get_nowait()
except:
sleep(0.1)
continue
else:
for chunk in data_gen:
queue.put(chunk)
## Setup thread and data generator
thread.start_new_thread(chunk_yielder, (chunk_queue,))
num_of_chunks = 100
data_gen = binarize(num_of_chunks)
queue.put(data_gen)
## Get data back and send away
while True:
try:
binary_chunk = queue.get_nowait()
except:
sleep(0.1)
continue
else:
socket.sendall(binary_chunk)
socket.recv(1000) #Get confirmation
However, I did not see and performance imporovement - it did not work faster.
I don't understand threads/processes too well, and my question is whether it is possible (at all and in Python) to gain from this type of separation, and what would be a good way to go about it, either with threads or processess (or any other way - async etc).
EDIT:
As far as I've come to understand -
Multirpocessing requires serializing any sent data, so I'm double-sending every computed data.
Sending via socket.send() should release the GIL
Therefore I think (please correct me if I am mistaken) that a threading solution is the right way. However I'm not sure how to do it correctly.
I know cython can release the GIL off of threads, but since one of them is just socket.send/recv, my understanding is that it shouldn't be necessary.

You have two options for running things in parallel in Python, either use the multiprocessing (docs) library , or write the parallel code in cython and release the GIL. The latter is significantly more work and less applicable generally speaking.
Python threads are limited by the Global Interpreter Lock (GIL), I won't go into detail here as you will find more than enough information online on it. In short, the GIL, as the name suggests, is a global lock within the CPython interpreter that ensures multiple threads do not modify objects, that are within the confines of said interpreter, simultaneously. This is why, for instance, cython programs can run code in parallel because they can exist outside the GIL.
As to your code, one problem is that you're running both the number crunching (binarize) and the socket.send inside the GIL, this will run them strictly serially. The queue is also connected very strangely, and there is a NameError but let's leave those aside.
With the caveats already pointed out by Jeremy Friesner in mind, I suggest you re-structure the code in the following manner: you have two processes (not threads) one for binarising the data and the other for sending data. In addition to those, there is also the parent process that started both children, and a queue connecting child 1 to child 2.
Subprocess-1 does number crunching and produces crunched data into a queue
Subprocess-2 consumes data from a queue and does socket.send
in code the setup would look something like
from multiprocessing import Process, Queue
work_queue = Queue()
p1 = Process(target=binarize, args=(100, work_queue))
p2 = Process(target=send_data, args=(ip, port, work_queue))
p1.start()
p2.start()
p1.join()
p2.join()
binarize can remain as it is in your code, with the exception that instead of a yield at the end, you add elements into the queue
def binarize(num_of_chunks, q):
''' Generator function that yields chunks of binary data. In reality it wouldn't be the same data'''
ordered_col_list = ('col1', 'col2')
columns = dict.fromkeys(ordered_col_list)
for i in range(num_of_chunks):
columns['col1'] = array.array('i', range(10000)).tostring()
columns['col2'] = array.array('f', [float(num) for num in range(10000)]).tostring()
data = b''.join((columns[col_name] for col_name in ordered_col_list))
q.put(data)
send_data should just be the while loop from the bottom of your code, with the connection open/close functionality
def send_data(ip, addr, q):
s = socket.create_connection((ip, addr))
while True:
try:
binary_chunk = q.get(False)
except:
sleep(0.1)
continue
else:
socket.sendall(binary_chunk)
socket.recv(1000) # Get confirmation
# maybe remember to close the socket before killing the process
Now you have two (three actually if you count the parent) processes that are processing data independently. You can force the two processes to synchronise their operations by setting the max_size of the queue to a single element. The operation of these two separate processes is also easy to monitor from the process manager on your computer top (Linux), Activity Monitor (OsX), don't remember what it's called under Windows.
Finally, Python 3 comes with the option of using co-routines which are neither processes nor threads, but something else entirely. Co-routines are pretty cool from a CS point of view, but a bit of a head scratcher at first. There is plenty of resources to learn from though, like this post on Medium and this talk by David Beazley.
Even more generally, you might want to look into the producer/consumer pattern, if you are not already familiar with it.

If you are trying to use concurrency to improve performance in CPython I would strongly recommend using multiprocessing library instead of multithreading. It is because of GIL (Global Interpreter Lock), which can have a huge impact on execution speed (in some cases, it may cause your code to run slower than single threaded version). Also, if you would like to learn more about this topic, I recommend reading this presentation by David Beazley. Multiprocessing bypasses this problem by spawning a new Python interpreter instance for each process, thus allowing you to take full advantage of multi core architecture.

Python threading with filehandling

Hello I have a program that looks through a range of data and finds anomalies in that data. To make my program faster I incorporated the use of threads (66 in total) now when my program finds the anomalies I would want it to write it to a file but however when i try to write to the file from within multiple threads it wont write.
class myThread(threading.Thread):
def __init__(self,arg1,arg2,lock,output):
threading.Thread.__init__(self)
self.arg1 = arg1
self.arg2 = arg2
self.lock = lock
self.file = output
def run(self):
# print "Starting " + self.name
main(self.arg1,self.arg2,self.lock,self.file)
# print "Exiting " + self.name
def main(START_IP,END_IP,lock,File):
# store found DNS servers
foundDNS=[]
# scan all the ip addresses in the range
for i0 in range(START_IP[0], END_IP[0]+1):
for i1 in range(START_IP[1], END_IP[1]+1):
for i2 in range(START_IP[2], END_IP[2]+1):
for i3 in range(START_IP[3], END_IP[3]+1):
# build ip addres
ipaddr=str(i0)+"."+str(i1)+"."+str(i2)+"."+str(i3)
print "Scanning "+ipaddr+"...",
# scan address
ret=ScanDNS(ipaddr, 10)
if ret==True:
foundDNS.append(ipaddr)
print "Found!"
lock.acquire()
File.write(ipaddr)
File.write("\n")
File.flush()
lock.release()
else:
print
file = open("file.txt","wb")
lock = threading.Lock()
thread1 = myThread(START_IP,END_IP,lock,)
thread1.start()
This uses my exact same MyThread class just with the required arguments for main to manipulate the data. If I run my code for about a minute as its scanning over DNS servers I should get maybe 20-30 DNS servers saved into a file but I generally get this:
FILE.TXT
2.2.1.2
8.8.8.8
31.40.40
31.31.40.40
31.31.41.41
I know for a fact (because I watched the scanning output) and that it hardly writes all of them. So why is some writing and some not?

I don't know why your code is not working, but I can hazard a guess that it is due to race conditions. Hopefully someone knowledgeable can answer that part of your question.
However, I've encountered a similar problem before, and I solved it by moving the file writing code to a single output thread. This thread read from a synchronized queue to which other threads pushed data to be written.
Also, if you happen to be working on a machine with multiple cores, then it's better to use multiprocess instead of threading. The latter only runs threads on a single core, while the former does not have this limitation.

instead of providing file - provide Queue. Spawn new thread to read from Queue and file write. Or Use Locks everywhere in print too because some treads can be deadlocked.

To avoid potential error or misuse for access file from multi-threads, you can try using logging to write down your result.
import logging
logger = logging.getLogger()
file_handler = logging.FileHandler()
formatter = #your formmat
file_handler.setFormatter(formatter)
logger.addHandler(file_handler)

Check the the documentation for File Objects:
File.flush() is not enough to ensure that your data is written on disk, add
os.fsync(File.fileno()) just after to make it happens.

Problems mixing threads/processes in python [duplicate]

I have a piece of code below that creates a few threads to perform a task, which works perfectly well on its own. However I'm struggling to understand why the print statements I call in my function do not execute until all threads complete and the print 'finished' statement is called. I would expect them to be called as the thread executes. Is there any simple way to accomplish this, and why does this work this way in the first place?
def func(param):
time.sleep(.25)
print param*2
if __name__ == '__main__':
print 'starting execution'
launchTime = time.clock()
params = range(10)
pool=multiprocessing.Pool(processes=100) #use N processes to download the data
_=pool.map(func,params)
print 'finished'

For python 3 you can now use the flush param like that:
print('Your text', flush=True)

This happens due to stdout buffering. You still can flush the buffers:
import sys
print 'starting'
sys.stdout.flush()
You can find more info on this issue here and here.

Having run into plenty of issues around this and garbled outputs (especially under Windows when adding colours to the output..), my solution has been to have an exclusive printing thread which consumes a queue
If this still doesn't work, also add flush=True to your print statement(s) as suggested by #Or Duan
Further, you may find the "most correct", but a heavy-handed approach to displaying messages with threading is to use the logging library which can wrap a queue (and write to many places asynchronously, including stdout) or write to a system-level queue (outside Python; availability depends greatly on OS support)
import threading
from queue import Queue
def display_worker(display_queue):
while True:
line = display_queue.get()
if line is None: # simple termination logic, other sentinels can be used
break
print(line, flush=True) # remove flush if slow or using Python2
def some_other_worker(display_queue, other_args):
# NOTE accepts queue reference as an argument, though it could be a global
display_queue.put("something which should be printed from this thread")
def main():
display_queue = Queue() # synchronizes console output
screen_printing_thread = threading.Thread(
target=display_worker,
args=(display_queue,),
)
screen_printing_thread.start()
### other logic ###
display_queue.put(None) # end screen_printing_thread
screen_printing_thread.stop()

Sending data through a socket from another thread does not work in Python

This is my 'game server'. It's nothing serious, I thought this was a nice way of learning a few things about python and sockets.
First the server class initialized the server.
Then, when someone connects, we create a client thread. In this thread we continually listen on our socket.
Once a certain command comes in (I12345001001, for example) it spawns another thread.
The purpose of this last thread is to send updates to the client.
But even though I see the server is performing this code, the data isn't actually being sent.
Could anyone tell where it's going wrong?
It's like I have to receive something before I'm able to send. So I guess somewhere I'm missing something
#!/usr/bin/env python
"""
An echo server that uses threads to handle multiple clients at a time.
Entering any line of input at the terminal will exit the server.
"""
import select
import socket
import sys
import threading
import time
import Queue
globuser = {}
queue = Queue.Queue()
class Server:
def __init__(self):
self.host = ''
self.port = 2000
self.backlog = 5
self.size = 1024
self.server = None
self.threads = []
def open_socket(self):
try:
self.server = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.server.bind((self.host,self.port))
self.server.listen(5)
except socket.error, (value,message):
if self.server:
self.server.close()
print "Could not open socket: " + message
sys.exit(1)
def run(self):
self.open_socket()
input = [self.server,sys.stdin]
running = 1
while running:
inputready,outputready,exceptready = select.select(input,[],[])
for s in inputready:
if s == self.server:
# handle the server socket
c = Client(self.server.accept(), queue)
c.start()
self.threads.append(c)
elif s == sys.stdin:
# handle standard input
junk = sys.stdin.readline()
running = 0
# close all threads
self.server.close()
for c in self.threads:
c.join()
class Client(threading.Thread):
initialized=0
def __init__(self,(client,address), queue):
threading.Thread.__init__(self)
self.client = client
self.address = address
self.size = 1024
self.queue = queue
print 'Client thread created!'
def run(self):
running = 10
isdata2=0
receivedonce=0
while running > 0:
if receivedonce == 0:
print 'Wait for initialisation message'
data = self.client.recv(self.size)
receivedonce = 1
if self.queue.empty():
print 'Queue is empty'
else:
print 'Queue has information'
data2 = self.queue.get(1, 1)
isdata2 = 1
if data2 == 'Exit':
running = 0
print 'Client is being closed'
self.client.close()
if data:
print 'Data received through socket! First char: "' + data[0] + '"'
if data[0] == 'I':
print 'Initializing user'
user = {'uid': data[1:6] ,'x': data[6:9], 'y': data[9:12]}
globuser[user['uid']] = user
print globuser
initialized=1
self.client.send('Beginning - Initialized'+';')
m=updateClient(user['uid'], queue)
m.start()
else:
print 'Reset receivedonce'
receivedonce = 0
print 'Sending client data'
self.client.send('Feedback: ' +data+';')
print 'Client Data sent: ' + data
data=None
if isdata2 == 1:
print 'Data2 received: ' + data2
self.client.sendall(data2)
self.queue.task_done()
isdata2 = 0
time.sleep(1)
running = running - 1
print 'Client has stopped'
class updateClient(threading.Thread):
def __init__(self,uid, queue):
threading.Thread.__init__(self)
self.uid = uid
self.queue = queue
global globuser
print 'updateClient thread started!'
def run(self):
running = 20
test=0
while running > 0:
test = test + 1
self.queue.put('Test Queue Data #' + str(test))
running = running - 1
time.sleep(1)
print 'Updateclient has stopped'
if __name__ == "__main__":
s = Server()
s.run()

I don't understand your logic -- in particular, why you deliberately set up two threads writing at the same time on the same socket (which they both call self.client), without any synchronization or coordination, an arrangement that seems guaranteed to cause problems.
Anyway, a definite bug in your code is you use of the send method -- you appear to believe that it guarantees to send all of its argument string, but that's very definitely not the case, see the docs:
Returns the number of bytes sent.
Applications are responsible for
checking that all data has been sent;
if only some of the data was
transmitted, the application needs to
attempt delivery of the remaining
data.
sendall is the method that you probably want:
Unlike send(), this method continues
to send data from string until either
all data has been sent or an error
occurs.
Other problems include the fact that updateClient is apparently designed to never terminate (differently from the other two thread classes -- when those terminate, updateClient instances won't, and they'll just keep running and keep the process alive), redundant global statements (innocuous, just confusing), some threads trying to read a dict (via the iteritems method) while other threads are changing it, again without any locking or coordination, etc, etc -- I'm sure there may be even more bugs or problems, but, after spotting several, one's eyes tend to start to glaze over;-).

You have three major problems. The first problem is likely the answer to your question.
Blocking (Question Problem)
The socket.recv is blocking. This means that execution is halted and the thread goes to sleep until it can read data from the socket. So your third update thread just fills the queue up but it only gets emptied when you get a message. The queue is also emptied by one message at a time.
This is likely why it will not send data unless you send it data.
Message Protocol On Stream Protocol
You are trying to use the socket stream like a message stream. What I mean is you have:
self.server = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
The SOCK_STREAM part says it is a stream not a message such as SOCK_DGRAM. However, TCP does not support message. So what you have to do is build messages such as:
data =struct.pack('I', len(msg)) + msg
socket.sendall(data)
Then the receiving end will looking for the length field and read the data into a buffer. Once enough data is in the buffer it can grab out the entire message.
Your current setup is working because your messages are small enough to all be placed into the same packet and also placed into the socket buffer together. However, once you start sending large data over multiple calls with socket.send or socket.sendall you are going to start having multiple messages and partial messages being read unless you implement a message protocol on top of the socket byte stream.
Threads
Even though threads can be easier to use when starting out they come with a lot of problems and can degrade performance if used incorrectly especially in Python. I love threads so do not get me wrong. Python also has a problem with the GIL (global interpreter lock) so you get bad performance when using threads that are CPU bound. Your code is mostly I/O bound at the moment, but in the future it may become CPU bound. Also you have to worry about locking with threading. A thread can be a quick fix but may not be the best fix. There are circumstances where threading is quite simply the easiest way to break some time consuming process. So do not discard threads as evil or terrible. In Python they are considered bad mainly because of the GIL, and in other languages (including Python) because of concurrency issues so most people recommend you to use multiple processes with Python or use asynchronous code. The subject of to use a thread or not is very complex as it depends on the language (way your code is run), the system (single or multiple processors), and contention (trying to share a resource with locking), and other factors, but generally asynchronous code is faster because it utilizes more CPU with less overhead especially if you are not CPU bound.
The solution is the usage of the select module in Python, or something similar. It will tell you when a socket has data to be read, and you can set a timeout parameter.
You can gain more performance by doing asynchronous work (asynchronous sockets). To turn a socket into asynchronous mode you simply call socket.settimeout(0) which will make it not block. However, you will constantly eat CPU spinning waiting for data. The select module and friends will prevent you from spinning.
Generally for performance you want to do as much asynchronous (same thread) as possible, and then expand with more threads that are also doing as much asynchronously as possible. However as previously noted Python is an exception to this rule because of the GIL (global interpreter lock) which can actually degrade performance from what I have read. If you are interesting you should try writing a test case and find out!
You should also check out the thread locking primitives in the threading module. They are Lock, RLock, and Condition. They can help multiple threads share data with out problems.
lock = threading.Lock()
def myfunction(arg):
with lock:
arg.do_something()
Some Python objects are thread safe and others are not.
Sending Updates Asynchronously (Improvement)
Instead of using a third thread only to send updates you could instead use the client thread to send updates by checking the current time with the last time an update was sent. This would remove the usage of a Queue and a Thread. Also to do this you must convert your client code into asynchronous code and have a timeout on your select so that you can at interval check the current time to see if an update is needed.
Summary
I would recommend you rewrite your code using asynchronous socket code. I would also recommend that you use a single thread for all clients and the server. This will improve performance and decrease latency. It would make debugging easier because you would have no possible concurrency issues like you have with threads. Also, fix your message protocol before it fails.