Multithreading freezes when using `thread.join()`

Question

I am trying to set up 3 thread and execute 5 tasks in a queue. The idea is that the threads will first run the first 3 tasks at the same time, then 2 threads finish the remaining 2. But the program seems freeze. I couldn't detect anything wrong with it.

from multiprocessing import Manager
import threading
import time
global exitFlag 
exitFlag = 0


class myThread(threading.Thread):
    def __init__(self, threadID, name, q):
        threading.Thread.__init__(self)
        self.threadID = threadID
        self.name = name
        self.q = q

    def run(self):
        print("Starting " + self.name)
        process_data(self.name, self.q)
        print("Exiting " + self.name)


def process_data(threadName, q):
    global exitFlag
    while not exitFlag:
        if not workQueue.empty():
            data = q.get()
            print("%s processing %s" % (threadName, data))
        else:
            pass
        time.sleep(1)
    print('Nothing to Process')


threadList = ["Thread-1", "Thread-2", "Thread-3"]
nameList = ["One", "Two", "Three", "Four", "Five"]
queueLock = threading.Lock()
workQueue = Manager().Queue(10)
threads = []
threadID = 1

# create thread
for tName in threadList:
    thread = myThread(threadID, tName, workQueue)
    thread.start()
    threads.append(thread)
    threadID += 1

# fill up queue
queueLock.acquire()
for word in nameList:
    workQueue.put(word)
queueLock.release()

# wait queue clear
while not workQueue.empty():
    pass
# notify thread exit
exitFlag = 1
# wait for all threads to finish
for t in threads:
    t.join()
print("Exiting Main Thread")

I don't know what happened exactly, but after I remove the join() part, the program is able to run just fun. What I don't understand is that exitFlag is supposed to have sent out the signal when the queue is emptied. So it seems somehow the signal was not detected by process_data()

Answer 1

There are multiple issues with your code. First of, threads in CPython don't run Python code "at the same time" because of the global interpreter lock (GIL). A thread must hold the GIL to execute Python bytecode. By default a thread holds the GIL for up to 5 ms (Python 3.2+), if it doesn't drop it earlier because it does blocking I/O. For parallel execution of Python code you would have to use multiprocessing.

You also needlessly use a Manager.Queue instead of a queue.Queue. A Manager.Queue is a queue.Queue on a separate manager-process. You introduced a detour with IPC and memory copying for no benefit here.

The cause of your deadlock is that you have a race condition here:

    if not workQueue.empty():
        data = q.get()

This is not an atomic operation. A thread can check workQueue.empty(), then drop the GIL, letting another thread drain the queue and then proceed with data = q.get(), which will block forever if you don't put something again on the queue. Queue.empty() checks are a general anti-pattern and there is no need to use it. Use poison pills (sentinel-values) to break a get-loop instead and to let the workers know they should exit. You need as many sentinel-values as you have workers. Find more about iter(callabel, sentinel) here.

import time
from queue import Queue
from datetime import datetime
from threading import Thread, current_thread


SENTINEL = 'SENTINEL'


class myThread(Thread):

    def __init__(self, func, inqueue):
        super().__init__()
        self.func = func
        self._inqueue = inqueue

    def run(self):
        print(f"{datetime.now()} {current_thread().name} starting")
        self.func(self._inqueue)
        print(f"{datetime.now()} {current_thread().name} exiting")


def process_data(_inqueue):
    for data in iter(_inqueue.get, SENTINEL):
        print(f"{datetime.now()} {current_thread().name} "
              f"processing {data}")
        time.sleep(1)


if __name__ == '__main__':


    N_WORKERS = 3

    inqueue = Queue()
    input_data = ["One", "Two", "Three", "Four", "Five"]

    sentinels = [SENTINEL] * N_WORKERS # one sentinel value per worker
    # enqueue input and sentinels
    for word in input_data +  sentinels:
        inqueue.put(word)

    threads = [myThread(process_data, inqueue) for _ in range(N_WORKERS)]

    for t in threads:
        t.start()
    for t in threads:
        t.join()

    print(f"{datetime.now()} {current_thread().name} exiting")

Example Output:

2019-02-14 17:58:18.265208 Thread-1 starting
2019-02-14 17:58:18.265277 Thread-1 processing One
2019-02-14 17:58:18.265472 Thread-2 starting
2019-02-14 17:58:18.265542 Thread-2 processing Two
2019-02-14 17:58:18.265691 Thread-3 starting
2019-02-14 17:58:18.265793 Thread-3 processing Three
2019-02-14 17:58:19.266417 Thread-1 processing Four
2019-02-14 17:58:19.266632 Thread-2 processing Five
2019-02-14 17:58:19.266767 Thread-3 exiting
2019-02-14 17:58:20.267588 Thread-1 exiting
2019-02-14 17:58:20.267861 Thread-2 exiting
2019-02-14 17:58:20.267994 MainThread exiting

Process finished with exit code 0

If you don't insist on subclassing Thread, you could also just use multiprocessing.pool.ThreadPool a.k.a. multiprocessing.dummy.Pool which does the plumbing for you in the background.