Implementation of singleflight-cluster.
By default, it uses consistent hashing on top of zookeeper + kazoo to handle service discovery and use falcon web framework for backend server.
All of the parameters are duck-typed, so you can look at the example implementation if you want to use another implementationn.
This implementation is eventually consistent, in which during any request lifetime,
the instances known to each one may be different, but the force_this_node paramater is used
on backend call path, so it would never live-locking, calling each other
believing the other is still the rightful owner.
This approach works because zookeeper will return a snapshot of instances, in which the requested node
might be the rightful owner of the key, one time in the past.
If disconnected from zookeeper for too long (in kazoo term, SUSPENDED or LOST), after a specified disconnected_timeout (default 60s), the instance will no longer be valid and start rejecting request, to prevent further membership drifting
See tests/core.py for example on how to initialize and use this implementation.