mutextree

To protect tree-like resources:

Nodes have to be represented by their fully qualified name. For instance, the node D is represented by [A, B, D]:

The algorithm to lock a node is the following:

1. We lock the parent nodes in order:

2. Then we check that no child node is already locked:

3. We lock the wanted node

4. We release all the parent locks that are no longer usefull.

Interface targeted to be exactly like threading.Lock.

The mutex tree is actually designed to use redis and python-redis-lock but the locking backend may be changed.

Usage

To use mutextree with the redis locks back end, simply instanciate a redis client and create your lock. The redis client should be strict and decode responses itself.

import redis
from mutextree import RedisLockBackend, TreeLock

redis_client = redis.StrictRedis(decode_responses=True)
redis_lock_backend = RedisLockBackend(redis_client)
lock = TreeLock(redis_lock_backend, ["nodeA"], expire=10, timeout=10, id=1)
try:
    lock.acquire()
    # do things
finally:
    lock.release()

You can use it as a context manager or a decorator:

import redis
from mutextree import RedisLockBackend, TreeLock, tree_lock

redis_client = redis.StrictRedis(decode_responses=True)
redis_lock_backend = RedisLockBackend(redis_client)

with TreeLock(redis_lock_backend, ["nodeA"], expire=10):
    # do things
    pass
    # release will be automatically called

# Or with a decorator
@tree_lock(redis_lock_backend, ["nodeA"], expire=10)
def protected_function():
    # do things
    pass
    # release will be automatically called

Lock has the same basic interface as threading.Lock() with some more methods: acquire, release, refresh.

Tests

Tests should be run under python 2.7 and python 3.6 to tests everything

$ pip install tox
$ tox -e py27,py36

Coverage reports will be the merge of the coverage for py27 and py36.