Groot provides an eventually consistent, ephemeral KV store. It relies on distributed erlang and uses LWW-registers and Hybrid-logical clocks to provide maximum availability. Groot utilizes ETS for efficient reading.

Usage

# Changes are propogated to other nodes.
:ok = Groot.set(:key, "value")

# Read existing values
"value" = Groot.get(:key)

Updates will replicate to all connected nodes. If new nodes join, or if a node rejoins the cluster after a network partition then the other nodes in the cluster will replicate all of their known registers to the new node.

Consistency

Groot provides LWW registers. Each register includes a hybrid logical clock (HLC). Ordering of events is determined by comparing HLCs. If a network partition occurs nodes on either side of the partition will continue to accept set and get operations. Once the partition heals all registers will be replicated to all nodes. If there are any conflicts the register with the largest HLC will be chosen.

Its possible that Groot will lose writes under specific failures scenarios. For instance, if there is a network partition between 2 nodes, neither node will be able to replicate to the other. If either node crashes after accepting a write, that write will be lost.

Data limitations

Groot replicates all keys to all connected nodes. Thus there may be performance issues if you attempt to store hundreds or thousands of keys. This issue may be fixed in a future release.