Building a Distributed Key-Value Store in C++ (Part 4)

With a working TCP server from Part 3, the next step is running multiple instances and keeping them in sync. This part adds a ClusterManager, a PeerCommunicator, and replication forwarding to KVServer.

ClusterManager - knowing the cluster topology

class ClusterManager {
public:
  explicit ClusterManager(const std::vector<std::pair<std::string, int>>& peers);
  const std::vector<std::pair<std::string, int>>& get_peers() const;
private:
  std::vector<std::pair<std::string, int>> peer_addresses;
};

Each server is started with a list of peer host:port pairs. ClusterManager holds that list and exposes it to the rest of the server code.

PeerCommunicator - fire and forget

class PeerCommunicator {
public:
  static void send_command(const std::string& host, int port, const std::string& cmd);
};

send_command opens a short-lived TCP connection, writes the command, and closes. There is no acknowledgement or retry - this is intentionally the simplest possible implementation.

Wiring replication into KVServer

Forwarding a PUT

if (cmd == "PUT") {
  store.put(key, value);
  forward_to_peers("REPL_PUT " + key + " " + value);
  response << "OK\n";
}

The store is updated locally first, then the write is forwarded. If forwarding fails silently, the local node still has the data - peers will be stale.

Iterating over peers

for (const auto& [host, port] : cluster.get_peers()) {
  PeerCommunicator::send_command(host, port, cmd);
}

REPL_PUT and REPL_DEL are handled as internal commands on the receiving side - they update the store without triggering another round of replication (avoiding infinite loops).

Launching a 3-node cluster

./kvstore_server 12345 127.0.0.1:12346,127.0.0.1:12347
./kvstore_server 12346 127.0.0.1:12345,127.0.0.1:12347
./kvstore_server 12347 127.0.0.1:12345,127.0.0.1:12346

Each node lists the other two as peers. A PUT on port 12345 propagates to 12346 and 12347 within a few milliseconds on localhost.

CMake library organisation

add_library(kvstore_lib
  main/kvstore.cpp
  core/memory_backend.cpp
  core/file_append_log.cpp
  net/cluster_manager.cpp
  net/peer_communicator.cpp
)

Grouping source files by concern (core/, net/) keeps the build file readable and makes it easier to swap implementations later (e.g., replacing file_append_log with an LSM-tree backend).

What's next

Part 5 adds leader election via the Bully algorithm so writes are coordinated through a single node instead of accepted by all.