An interactive study companion for Martin Kleppmann's classic on reliable, scalable, and maintainable systems.
Consensus protocols let nodes agree on a value despite failures — powering distributed transactions.
Getting multiple nodes to agree on a single value — who is leader, whether to commit, what the next log entry is — is the consensus problem. Algorithms like Raft make this practical for production systems.
etcd and HashiCorp Consul use Raft for leader election and consistent key-value storage — the backbone of Kubernetes control planes. Kafka replaced ZooKeeper with KRaft (internal Raft) for broker metadata. Cross-database 2PC (XA transactions) exists in Java EE but is fragile; the saga pattern with compensating transactions in microservices is more common today.
Chubby (precursor to etcd) provided distributed consensus for Google's internal infrastructure — leader election, lock service, and membership. Kubernetes today relies on etcd's Raft implementation for the same guarantees.
// Microservice SAGA — compensate if payment fails (Airbnb-style)
async function bookListing(input: BookingInput) {
const hold = await inventoryService.reserve(input); // step 1
try {
const payment = await paymentService.charge(input); // step 2
await notificationService.sendConfirmation(input); // step 3
return { hold, payment };
} catch (err) {
await inventoryService.release(hold.id); // compensating transaction
throw err;
}
}