SWIM Protocol

This is an implementation of the SWIM protocol in Java, which stands for "Scalable Weakly-consistent Infection-style Process Group Membership Protocol". See the original paper.

A clustered system needs a way to keep track of which of its members are alive and which have failed. One way to do this is through a group membership list, which is a list of the live members in a cluster. A copy of this list is available at each node and is kept synchronized in an eventually-consistent manner.

The SWIM protocol is a group membership list with two main components, a failure detector that is used to detect the alive/dead status of other members, and a dissemination component that is used to propagate this information to other nodes in the cluster.

Simulation Example

One of the goals of this project was to be able to swap out the UDP network layer with an emulated network layer. This enables the ability to run multi-node simulations quickly and in a single JVM process.

Here is an example of a simulation:

voidtestWithPacketLossAndNoFailures() throwsExecutionException, InterruptedException {
// A 5% chance that the emulated network will drop a message is added, allowing us to test the robustness of the protocol.doubledropProbability = 0.05;
intmaxRounds = 30;
intnode1Port = 5555;
intnode2Port = 5556;
intnode3Port = 5557;
SimulationQueuessimulationQueues = newSimulationQueues();
// The nodes join the cluster on rounds 1, 3 and 8 respectively.SimulationNodenode1 = newSimulationNode.Builder()
.withJoinTime(1)
.withDropProbability(dropProbability)
.build(node1Port, simulationQueues);
SimulationNodenode2 = newSimulationNode.Builder()
.withJoinTime(3)
.withDropProbability(dropProbability)
.build(node2Port, simulationQueues);
SimulationNodenode3 = newSimulationNode.Builder()
.withJoinTime(8)
.withDropProbability(dropProbability)
.build(node3Port, simulationQueues);
List<SimulationNode> nodes = Arrays.asList(node1, node2, node3);
// Creates a 3 node simulation that runs the protocol 30 times for each node.Simulationsimulation = newSimulation(nodes, maxRounds);
simulation.run();
// The simulation ends by ensuring that each member has a complete copy of the list, in spite of the "network jitter" that we introduced.for (SimulationNodenode : nodes) {
assertEquals(nodes.size(), node.getMemberList().size());
}
}

Building

To compile

./gradlew build

To run unit tests

./gradlew test

To run simulations

./gradlew integrationTest