I deal with two python queues.
Short description of my issue:
Clients pass through the waiting queue
(q1) and they (the clients) are served afterwards.
The size of the waiting queue
can't be greater than N (10 in my program).
If waiting queue
becomes full, clients pass to outside queue
(q2, size 20). If outside queue becomes full, clients are rejected and not served.
Every client that left a waiting queue allows another client from outside queue to join the waiting queue.
Work with queues should be thread-safe.
Below I implemented approximately what I want. But I'm faced with the problem - enqueuing a client from outside queue (q1) to the waiting queue (q2) during execution serve
function. I guess I lost or forgot something important. I think this statement q1.put(client)
blocks permanently but don't know why.
import time
import threading
from random import randrange
from Queue import Queue, Full as FullQueue
class Client(object):
def __repr__(self):
return '<{0}: {1}>'.format(self.__class__.__name__, id(self))
def serve(q1, q2):
while True:
if not q2.empty():
client = q2.get()
print '%s leaved outside queue' % client
q1.put(client)
print '%s is in the waiting queue' % client
q2.task_done()
client = q1.get()
print '%s leaved waiting queue for serving' % client
time.sleep(2) # Do something with client
q1.task_done()
def main():
waiting_queue = Queue(10)
outside_queue = Queue(20)
for _ in range(2):
worker = threading.Thread(target=serve, args=(waiting_queue, outside_queue))
worker.setDaemon(True)
worker.start()
delays = [randrange(1, 5) for _ in range(100)]
# Every d seconds 10 clients enter to the waiting queue
for d in delays:
time.sleep(d)
for _ in range(10):
client = Client()
try:
waiting_queue.put_nowait(client)
except FullQueue:
print 'Waiting queue is full. Please line up in outside queue.'
try:
outside_queue.put_nowait(client)
except FullQueue:
print 'Outside queue is full. Please go out.'
waiting_queue.join()
outside_queue.join()
print 'Done'
Finally I found the solution. I check docs more attentive
If full() returns True it doesn’t guarantee that a subsequent call to get() will not block
https://docs.python.org/2/library/queue.html#Queue.Queue.full
That's why q1.full()
is not reliable in a few threads. I added mutex before inserting item to queues and checking queue is full:
class Client(object):
def __init__(self, ident):
self.ident = ident
def __repr__(self):
return '<{0}: {1}>'.format(self.__class__.__name__, self.ident)
def serve(q1, q2, mutex):
while True:
client = q1.get()
print '%s leaved waiting queue for serving' % client
time.sleep(2) # Do something with client
q1.task_done()
with mutex:
if not q2.empty() and not q1.full():
client = q2.get()
print '%s leaved outside queue' % client
q1.put(client)
print '%s is in the waiting queue' % client
q2.task_done()
def main():
waiting_queue = Queue(10)
outside_queue = Queue(20)
lock = threading.RLock()
for _ in range(2):
worker = threading.Thread(target=serve, args=(waiting_queue, outside_queue, lock))
worker.setDaemon(True)
worker.start()
# Every 1-5 seconds 10 clients enter to the waiting room
i = 1 # Used for unique <int> client's id
while True:
delay = randrange(1, 5)
time.sleep(delay)
for _ in range(10):
client = Client(i)
try:
lock.acquire()
if not waiting_queue.full():
waiting_queue.put(client)
else:
outside_queue.put_nowait(client)
except FullQueue:
# print 'Outside queue is full. Please go out.'
pass
finally:
lock.release()
i += 1
waiting_queue.join()
outside_queue.join()
print 'Done'
Now it works well.