core: added mutex lock

[alejandromumo]

closes inveniosoftware/invenio-users-resources#102

[alejandromumo]

It seems that tests are not enabled in the CI. I ran them locally:

---------- coverage: platform darwin, python 3.9.16-final-0 ----------
Name                          Stmts   Miss  Cover   Missing
-----------------------------------------------------------
invenio_cache/__init__.py         8      0   100%
invenio_cache/_compat.py          5      0   100%
invenio_cache/bccache.py          8      0   100%
invenio_cache/config.py           6      0   100%
invenio_cache/decorators.py      31      0   100%
invenio_cache/errors.py           8      1    88%   21
invenio_cache/ext.py             31      0   100%
invenio_cache/lock.py            67      4    94%   72, 80, 84, 88
invenio_cache/proxies.py          7      0   100%
-----------------------------------------------------------
TOTAL                           171      5    97%

======================================================================================================================================== 17 passed, 9 skipped, 14 warnings in 0.40s ========================================================================================================================================

Uncovered lines for lock.py are just a property and three base methods that raise NotImplementedError

For errors.py it is a __str__ method that is not being tested.

[ntarocco]

I have re-enabled the CI tests

[ntarocco]

If the lock was not acquired for some reasons, we might want to return True or to fail:

Suggested change

	if self.acquired:
	success = self._cache.delete(self.lock_id)
	if self.acquired:
	success = self._cache.delete(self.lock_id)
	else:
	success = True

Otherwise, it might return False and you don't know what happened.
WDYT?

[ntarocco]

Re-iterating on this: I would remove all if self.aquired. You need to be sure to pass the CachedMutex instance to release. I would rely exclusively on the existence of the cache key instead.
What is the use case? Multiprocessing context?

My worry is the following: if improperly used, it might happen that I acquire in one process (one web node), and try to release on another process (another web node) later one. This will fail, unless the same instance of CachedMutex is shared.

[alejandromumo]

I agree with that.

However, the lock might be released by any thread right now.

E.g.

# Thread 1
lock = Mutex("id1")
lock.acquire()

# Thread 2
lock = Mutex("id1")
lock.release()

We could "sign" these locks with a unique ID (e.g. set the value to a UUID) and only allow the release / renewal if the thread passes the same uuid. We briefly talked about it but did not elaborate on it.