Using the flufl.lock library

The flufl.lock package provides NFS-safe file locking with timeouts for POSIX systems. The implementation is influenced by the GNU/Linux open(2) manpage, under the description of the O_EXCL option:

[...] O_EXCL is broken on NFS file systems, programs which rely on it for performing locking tasks will contain a race condition. The solution for performing atomic file locking using a lockfile is to create a unique file on the same fs (e.g., incorporating hostname and pid), use link(2) to make a link to the lockfile. If link() returns 0, the lock is successful. Otherwise, use stat(2) on the unique file to check if its link count has increased to 2, in which case the lock is also successful.

The assumption made here is that there will be no outside interference, e.g. no agent external to this code will ever link() to the specific lock files used.

Lock objects support lock-breaking so that you can’t wedge a process forever.

Locks have a lifetime, which is the maximum length of time the process expects to retain the lock. It is important to pick a good number here because other processes will not break an existing lock until the expected lifetime has expired. Too long and other processes will hang; too short and you’ll end up trampling on existing process locks – and possibly corrupting data. In a distributed (NFS) environment, you also need to make sure that your clocks are properly synchronized.

Creating a lock

To create a lock, you must first instantiate a Lock object, specifying the path to a file that will be used to synchronize the lock. This file should not exist.

# This function comes from the test infrastructure.
>>> filename = temporary_lockfile()

>>> from flufl.lock import Lock
>>> lock = Lock(filename)
>>> lock
<Lock ... [unlocked: 0:00:15] pid=... at ...>

Locks have a default lifetime...

>>> lock.lifetime
datetime.timedelta(0, 15)

...which you can change.

>>> from datetime import timedelta
>>> lock.lifetime = timedelta(seconds=30)
>>> lock.lifetime
datetime.timedelta(0, 30)
>>> lock.lifetime = timedelta(seconds=15)

You can ask whether the lock is acquired or not.

>>> lock.is_locked
False

Acquiring the lock is easy if no other process has already acquired it.

>>> lock.lock()
>>> lock.is_locked
True

Once you have the lock, it’s easy to release it.

>>> lock.unlock()
>>> lock.is_locked
False

It is an error to attempt to acquire the lock more than once in the same process.

>>> from flufl.lock import AlreadyLockedError
>>> lock.lock()
>>> try:
...     lock.lock()
... except AlreadyLockedError as error:
...     print(error)
We already had the lock

>>> lock.unlock()

Lock objects also support the context manager protocol.

>>> lock.is_locked
False
>>> with lock:
...     lock.is_locked
True
>>> lock.is_locked
False

Lock acquisition blocks

When trying to lock the file when the lock is unavailable (because another process has already acquired it), the lock call will block.

>>> import time
>>> t0 = time.time()

# This function comes from the test infrastructure.
>>> acquire(filename, timedelta(seconds=5))
>>> lock.lock()
>>> t1 = time.time()
>>> lock.unlock()

>>> t1 - t0 > 4
True

Refreshing a lock

A process can refresh a lock if it realizes that it needs to hold the lock for a little longer. You cannot refresh an unlocked lock.

>>> from flufl.lock import NotLockedError
>>> try:
...     lock.refresh()
... except NotLockedError as error:
...     print(error)
<Lock ...

To refresh a lock, first acquire it with your best guess as to the length of time you’ll need it.

>>> from datetime import datetime
>>> lock.lifetime = timedelta(seconds=2)
>>> lock.lock()
>>> lock.is_locked
True

After the current lifetime expires, the lock is stolen from the parent process even if the parent never unlocks it.

# This function comes from the test infrastructure.
>>> t_broken = waitfor(filename, lock.lifetime)
>>> lock.is_locked
False

However, if the process holding the lock refreshes it, it will hold it can hold it for as long as it needs.

>>> lock.lock()
>>> lock.refresh(timedelta(seconds=5))
>>> t_broken = waitfor(filename, lock.lifetime)
>>> lock.is_locked
False

Lock details

Lock files are written with unique contents that can be queried for information about the host name the lock was acquired on, the id of the process that acquired the lock, and the path to the lock file.

>>> import os
>>> lock.lock()
>>> hostname, pid, lockfile = lock.details
>>> hostname == lock.hostname
True
>>> pid == os.getpid()
True
>>> lockfile == filename
True
>>> lock.unlock()

Even if another process has acquired the lock, the details can be queried.

>>> acquire(filename, timedelta(seconds=3))
>>> lock.is_locked
False
>>> hostname, pid, lockfile = lock.details
>>> hostname == lock.hostname
True
>>> pid == os.getpid()
False
>>> lockfile == filename
True

However, if no process has acquired the lock, the details are unavailable.

>>> lock.lock()
>>> lock.unlock()
>>> try:
...     lock.details
... except NotLockedError as error:
...     print(error)
Details are unavailable

Lock file separator

Lock claim file names contain useful bits of information concatenated by a separator character. This character is the caret (^) by default on Windows and the vertical bar (|) by default everywhere else. You can change this character. There are some restrictions:

  • It cannot be an alphanumeric;
  • It cannot appear in the host machine’s fully qualified domain name (e.g. the value of lock.hostname);
  • It cannot appear in the lock’s file name (the argument passed to the Lock constructor)

It may also be helpful to avoid any reserved characters on the file systems where you intend to run the code.

>>> lock = Lock(filename, separator='+')
>>> lock.lock()
>>> hostname, pid, lockfile = lock.details
>>> hostname == lock.hostname
True
>>> pid == os.getpid()
True
>>> lockfile == filename
True
>>> with open(filename) as fp:
...     claim_file = fp.read().strip()
...     '+' in claim_file
True
>>> lock.unlock()