Multi-Paxos Distributed Banking System

A fault-tolerant distributed banking system built on Multi-Paxos, supporting continuous transaction processing with leader election, crash fault tolerance, and exactly-once semantics.

This project implements a stable-leader Paxos variant to efficiently replicate a sequence of transactions across replicas while tolerating node failures.

---

Features

• Multi-Paxos with stable leader optimization
• Automatic leader election and recovery
• Exactly-once client semantics using timestamps
• Crash fault tolerance (fail-stop model)
• Deterministic total order execution
• Distributed key-value datastore (bank balances)
• Client retry and leader redirection logic
• Debugging & observability via introspection commands

---

System Architecture

• Nodes: 5 replicas (2f + 1, f = 2)
• Clients: 10 logical clients
• Replication: State machine replication
• Fault Model: Crash failures
• Communication: RPC / TCP-based message passing
• Execution Model: Out-of-order proposal, in-order execution

Each node maintains:

• Accepted log
• Commit log
• Execution pointer
• Client reply cache (for exactly-once semantics)

---

Protocol Overview

Normal Operation

1. Client sends (sender, receiver, amount) to leader
2. Leader assigns a monotonically increasing sequence number
3. Leader broadcasts ACCEPT(b, s, txn)
4. Replicas respond with ACCEPTED
5. Leader commits after majority and broadcasts COMMIT
6. All nodes execute transaction in order

Leader Election

• Triggered by timeout
• Uses PREPARE / PROMISE
• Accept logs are merged to avoid lost updates
• Missing sequence numbers are filled with no-op

---

Supported Commands

• PrintDB() — Print balances on all nodes
• PrintLog(node) — Inspect Paxos logs
• PrintStatus(seq) — View status (Accepted / Committed / Executed)
• PrintView() — Show leader changes

---

Running the System

# Build
make

# Run nodes
./node --id 1
./node --id 2
...

# Run client
./client input.csv