A single process, persistent multi-producer, multi-consumer queue.

producer-consumer, python
pip install pqueue==0.1.7



pqueue is a simple persistent (disk-based) FIFO queue for Python.

pqueue goals are speed and simplicity. The development was initially based on the Queuelib code. Entries are saved on disk using pickle.


  • Python 2.7 or Python 3.x
  • no external libraries requirements


You can install pqueue either via Python Package Index (PyPI) or from source.

To install using pip:

$ pip install pqueue

To install using easy_install:

$ easy_install pqueue

If you have downloaded a source tarball you can install it by running the following (as root):

# python setup.py install

How to use

pqueue provides a single FIFO queue implementation.

Here is an example usage of the FIFO queue:

>>> from pqueue import Queue
>>> q = Queue("tmpqueue")
>>> q.put(b'a')
>>> q.put(b'b')
>>> q.put(b'c')
>>> q.get()
>>> del q
>>> q = Queue("tmpqueue")
>>> q.get()
>>> q.get()
>>> q.get_nowait()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python2.7/Queue.py", line 190, in get_nowait
    return self.get(False)
  File "/usr/lib/python2.7/Queue.py", line 165, in get
    raise Empty

The Queue object is identical to Python's Queue module (or queue in Python 3.x), with the difference that it requires a parameter path indicating where to persist the queue data and chunksize indicating how many enqueued items should be stored per file. The same maxsize parameter available on the system wise Queue has been maintained.

In other words, it works exactly as Python's Queue, with the difference any abrupt interruption is ACID-guaranteed:

q = Queue()

def worker():
    while True:
        item = q.get()

for i in range(num_worker_threads):
     t = Thread(target=worker)
     t.daemon = True

for item in source():

q.join()       # block until all tasks are done

Note that pqueue is not intended to be used by multiple processes.

How it works?

Pushed data is serialized using pickle in sequence, on chunked files named as qNNNNN, with a maximum of chunksize elements, all stored on the given path.

The queue is formed by a head and a tail. Pushed data goes on head, pulled data goes on tail.

An info file is pickled in the path, having the following dict:

  • head: a list of three integers, an index of the head file, the number of elements written, and the file position of the last write.
  • tail: a list of three integers, an index of the tail file, the number of elements read, and the file position of the last read.
  • size: number of elements in the queue.
  • chunksize: number of elements that should be stored in each disk queue file.

Both read and write operations depend on sequential transactions on disk. In order to accomplish ACID requirements, these modifications are protected by the Queue locks.

If, for any reason, the application stops working in the middle of a head write, a second execution will remove any inconsistency by truncating the partial head write.

On get, the info file is not updated, only when you first call task_done, and only on the first time case you have to call it sequentially.

The info file is updated in the following way: a temporary file (using 'mkstemp') is created with the new data and then moved over the previous info file. This was designed this way as POSIX 'rename' is guaranteed to be atomic.

In case of abrupt interruptions, one of the following conditions may happen:

  • A partial write of the last pushed element may occur and in this case only this last element pushed will be discarded.
  • An element pulled from the queue may be processing, and in this case a second run will consume same element again.


Tests are located in pqueue/tests directory. They can be run using Python's default unittest module with the following command:


The output should be something like the following:

test_GarbageOnHead (pqueue.tests.test_queue.PersistenceTest)
Adds garbage to the queue head and let the internal integrity ... ok
test_MultiThreaded (pqueue.tests.test_queue.PersistenceTest)
Create consumer and producer threads, check parallelism ... ok
test_OpenCloseOneHundred (pqueue.tests.test_queue.PersistenceTest)
Write 1000 items, close, reopen checking if all items are there ... ok
test_OpenCloseSingle (pqueue.tests.test_queue.PersistenceTest)
Write 1 item, close, reopen checking if same item is there ... ok
test_PartialWrite (pqueue.tests.test_queue.PersistenceTest)
Test recovery from previous crash w/ partial write ... ok
test_RandomReadWrite (pqueue.tests.test_queue.PersistenceTest)
Test random read/write ... ok

Ran 6 tests in 1.301s



This software is licensed under the BSD License. See the LICENSE file in the top distribution directory for the full license text.


This software follows Semantic Versioning