Skip to content

Simple client for distributed locking built on top of python-consul.

License

Notifications You must be signed in to change notification settings

kurtome/python-consul-lock

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

52 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

¡ NOTICE !

This project is not being actively maintained, but it should serve as a good reference point for anyone interested in the same creating a fluent Python API for locking in Consul. (Just watch out for this issue.) Feel free to fork, or let me know if you are interested in taking over the project and maintaining it.

Python Consul Lock

Read this issue before using! #4

Simple client for distributed locking built on top of python-consul.

When running app servers in parallel distributed locks come in handy on the rare occasion you need guarentees that only one server is running a particular block of code at the same time. This library lets you do that in a straightforward way using Consul as the central authority for who owns the lock currently.

Installation

pip install consul-lock

Ephemeral Lock

Designed for relatively short-lived use-cases, primarily preventing race-conditions in application logic hot-spots. Locks are single use! The lock guarantees that no other client has locked that key concurrently.

Usable with lock/release in a try/finally block, or more easily via the the hold method in a with block. By default acquiring the lock is assumed to be a critical path, and will throw an exception if unable to acquire. The lock has a maximum (configurable) lifespan, which can prevent deadlocks or stale locks in the event that a lock is never released due to code crashes.

No guarentees are made about the behavior if a client continues to hold the lock for longer than its maximum lifespan (lock_timeout_seconds), Consul will release the lock at some point soon after the timeout. This is a good in thing, it is in fact the entire point of an ephemeral lock, because it makes it nearly impossible for stale locks to gum up whatever you are processing. The ideal setup if to configure the lock_timeout_seconds to be just long enough that there is no way your critical block could still be running, so it's safe enough to assume that the code that originally acquired the lock simply died.

The ephemeral lock is implemented with Consul's session and kv API and the key/value associated with the lock will be deleted upon release.

Examples

Setup consul_lock defaults

In order to create a lock, you must either pass in a reference to a consul.Consul client each time, or assign a default client to use.

import consul
import consul_lock

consul_client = consul.Consul()
consul_lock.defaults.consul_client = consul_client
Creating and holding a lock with as a context manager

The simplest way to use a lock is in a with block as a context manager. The lock will be automatically released then the with block exits.

from consul_lock import EphemeralLock

ephemeral_lock = EphemeralLock('my/special/key', acquire_timeout_ms=500)
with ephemeral_lock.hold():
    # do dangerous stuff here
    print 'here be dragons'
Creating a lock and acquiring and releasing it explicitly

It is also possible to manually acquire and release the lock. The following is equivalent to the previous example.

from consul_lock import EphemeralLock

ephemeral_lock = EphemeralLock('my/special/key', acquire_timeout_ms=500)
try:
    ephemeral_lock.acquire()
    # do dangerous stuff here
    print 'here be dragons'
finally:
    ephemeral_lock.release()
Reacting to acquire attempt

By default acquiring a lock (with acquire or hold) is assumed to be a critical operation and will throw an exception if it is unable to acquire the lock within the specified timeout. Sometimes it may be desirable to react to the fact that the lock is being held concurrently by some other code or host. In that case you can set the fail_hard option and acquire will return whether or not is was able to acquire the lock.

from consul_lock import EphemeralLock

ephemeral_lock = EphemeralLock('my/special/key', acquire_timeout_ms=500)
try:
    was_acquired = ephemeral_lock.acquire(fail_hard=False)
    if was_acquired:
        # do dangerous stuff here
        print 'here be dragons'
    else:
        print 'someone else has the lock :\ try again later'
finally:
    ephemeral_lock.release()

Lock configuration

Most of these settings can be both configured in consul_locks.defaults and overridden on each creation of the lock as keyword argments to the lock class.

  • consul_client - The instance of consul.Consul to use for accessing the Consul API. (no default, must be set or overridden)

  • acquire_timeout_ms - How long, in milliseconds, the caller is willing to wait to acquire the lock. When set to 0 lock acquisition will fail if the lock cannot be acquired immediately. (default = 0)

  • lock_timeout_seconds - How long, in seconds, the lock will stay alive if it is never released, this is controlled by Consul's Session TTL and may stay alive a bit longer according to their docs. As of the current version of Consul, this must be between 10 and 86400. (default = 180)

  • lock_key_pattern - A format string which will be combined with the key parameter for each lock to determine the full key path in Consul's key/value store. Useful for setting up a prefix path which all locks live under. This can only be set in consul_locks.defaults. (default = 'locks/ephemeral/%s')

  • generate_value - This can only be set in the consul_locks.defaults. (defaults to a function returning a JSON string containing "locked_at": str(datetime.now()))

FAQ

Is this "production ready"?

Use at your own risk, the locks the Consul supports via it's Sessions and Key/Value store weren't meant to be used for short lived locks, see this issue for more details. Test it out in your own setup with your expected usage pattern before using in a production system!

Why is this useful?

Well, that really depends on what you're doing, but generally distributed locks are useful to prevent race conditions.

How should I choose my key when locking?

Lock keys should be a specific as possible to the critical block of code the lock is protecting.

For example, one use case of locking may be to prevent emailing a welcome email upon signing up for a service.:

  • "send/email" - this is a terrible key to lock on, because it would affect all user emails across your entire code base. You would only be able to send one email at a time!
  • "send/user-123456/welcome-email" - assuming that the "123456" part is the user's ID, this is actually a pretty good lock because if user "123457" signs up at the exact same time, no problem! The locks for each user are unique, and can be acquired concurrently.
Ephemeral?!

So, you may be asking yourself, "I just double checked the definition for ephemeral, and dissapearing locks doen't sound too safe...wtf?" There is something to be said for not being too safe, if locks never dissapeared then what would happen if a chaos monkey came in and unplugged the server that acquired the lock? It would never be released, and you'd have to go in by hand and delete the lock in order to run your critical block of code.

Is the lock reentrant?

Nope, so be careful not to deadlock! If you somehow try to lock the same key while already holding a lock on that key, it will always fail until something times out.

Reentrant locking could be implemented since Consul's session API allows the same session to reacquire the same locked key, feel free to submit a pull request if you want that.

Has anyone actually asked any of these questions?

Nope.

Testing

Execute the following commands in the root directory of this project to run the tests.

Unit tests
python -m unittest -v consul_lock.tests.tests
Integration tests

These tests need to actually connect to a Consul cluster and read/write data. Some of these are slow due to testing of timeouts.

CONSUL_LOCK_CONSUL_HOST="127.0.0.1" CONSUL_LOCK_CONSUL_PORT=8500 python -m unittest -v consul_lock.integration_tests.tests

About

Simple client for distributed locking built on top of python-consul.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages