JustinDB

Distributed Key-Value Storage built on top of Scala/Akka

JustinDB KV is an eventually consistent key-value database that favours write availability. It’s a faithful implementation of Amazon’s Dynamo, with advanced features such as vector clocks for conflict resolution. JustinDB is also fault-tolerant. Servers can go up or down at any moment with no single point of failure.

Summary of techniques

Problem	Technique	Advantage
Partitioning	Consistent Hashing	Incremental Scalability
Membership and failure detection	Gossip-based membership protocol and failure detection	Preserves symmetry and avoids having a centralized registry for storing membership and node liveness information
High Availability for writes	Vector clocks with reconciliation during reads	Version size is decoupled from update rites
Recovering from permanent failures	Anti-entropy using Merkle trees	Synchronizes divergent replicas in the background
Conflicts resolution	CRDTs	Automatic conflict resolution during reads

Why akka

Its a toolkit and runtime for building highly concurrent applications which comes with ideas that have been around from some time - actor model. Besides that it has many welcome features around clustering:

1. load balancing
2. location transparency
3. self maintenance
4. fault tolerance

Authentication, authorization, validation

In case it's not obvious, Justin performs no authentication, authorization, or any validation of input data. Clients must implement those things themselves.

System Requirements

JustinDB works with Java 8 and newer.