CRDTs in Python

Heavily inspired by Meangirls

All sets support add, remove, merge, clone, to_dict, to_list, to_json methods and a from_dict class method.

Following examples use a hypothetical set up of 3 replicas which receive requests to add or remove an element and a cache layer that merges the results from all the replicas before returning a read response.

>>> import crdt
# Replica A
>>> a = crdt.TwoPhaseSet() # or crdt.LWWSet() or crdt.ORSet()
# Replica B
>>> b = crdt.TwoPhaseSet() # or crdt.LWWSet() or crdt.ORSet()
# Replica C
>>> c = crdt.TwoPhaseSet() # or crdt.LWWSet() or crdt.ORSet()

>>> a.add('foo')
>>> b.add('bar')
>>> c.remove('foo') # TwoPhaseSet will raise an exception here

# Cache layer that merges all replicas before returning the read response. 
# To do this over the wire:
# 1) to_json() to serialize
# 2) json.loads() to deserialize
# 3) and from_dict() to instantiate object again
>>> d = a.clone()

>>> d.merge(b)
>>> d.merge(c)

>>> d.to_list()
['foo', 'bar'] # if d is TwoPhaseSet
['bar']        # if d is LWWSet
['foo', 'bar'] # if d is ORSet

$ python -m unittest discover -s tests -p '*_test.py'

• The paper that introduced CRDTs and the morning paper entry on it
• Readings in conflict-free replicated data types should cover pretty much everything else in terms of theory.
• If you prefer watching talks like me:
  • Consistency without consensus in production systems by Peter Bourgon is the talk that got me excited about CRDTs because of their practical applications.
  • CRDTs Illustrated by Arnout Engelen also acts as a good companion to the paper.
  • In Practical Demystification of CRDT, Dmitry Ivanov & Nami Naserazad discuss pros & cons of a few of the CRDTs and some practical implementation issues.
• Other CRDT projects:
  • Meangirls in Ruby
  • Knockbox in Clojure
  • Akka CRDT for Akka
  • Roshi An LWW-element-set based implementation by SoundCloud