A master-master node cluster which replicates data.
A write is considered successful if k+1 nodes have accepted the data in memory, where k is maximum simultaneous failures. The writer itself is included in counting of maximum simultaneous failures.
A read is considered valid if 1 node has the value, plus k nodes are either unresponsive to read requests (they are failed or re-synchronizing), or have the value present.
fomostore is intended to be simple and implemented several times with diverging methods, so that a heterogenous cluster has minimal chances for simultaneous failure, and thus can more realistically accept a write without having sync'd the data to disk.
A related project aliclark/sensibility is intended to help understand how diverse a cluster's makeup is.
This distributed system is intended to be used as a building block, along with vanilla Raft, for building a distributed system.
Run your application code in duplicate on nodes in 2 or more availability zones (ensure that all effectful operations are idempotent and at-most-once), and use Raft to help decide what to do next, based on the data present in fomostore.