Trabica

Raft Consensus Algorithm implementation in Scala. This is pretty much still Work in progress.

Modules

Implemented features:

• Leader election
• Log replication (pluggable with a sample implementation)
• Persistence (pluggable with a sample implementation)
• Membership changes

A sample node initialization and a persistent log store is under opt/node-template

There are several modules currently implemented under the mod directory:

• lib-trabica-model Data models for node states and auxiliary data
• lib-trabica-net Abstraction for client and server networking
• lib-trabica-node Core Raft implementation
• lib-trabica-proto Protobuf model of client and server messages
• lib-trabica-rpc Protobuf service specification, along client and server endpoint factories
• lib-trabica-store Abstraction for pluggable Log store, a sample implementation is under opt/node-template

The code uses cats-effect and fs2 extensively.

Building

It is possible to build both a JVM and a native application. From SBT prompt run project node-template, then:

• stage for a JVM build, or
• native for a native image build

Running

Several shell scripts can be found under bin to start and experiment with a cluster, if you are initializing a new cluster, start with bin/bootstrap-00.sh. This will start a new cluster with a fresh log store, and expects 2 other nodes to join. bin/node-{n}.sh will start instances of the node in follower mode, where n is the node id.

To test membership changes, see the scripts under etc. Those scripts use grpcurl to send gRPC requests.

Configuration of election and RPC timeouts is under mod/lib-trabica-node/src/main/resources/reference.conf, which you can override in opt/node-template/src/main/resources/application.conf

If you start the example node with the default scripts under bin, logs are written under var/{n} directory, where n is the node id.

Notes

• The store implementation is just an example, a more robust store is needed in production situations
• A call to RemoveServer will aggressively send a termination signal to this server if removal was successful, which will terminate the node process
• Delivering (applying) messages to the FSM is not implemented yet

License

MIT