Skip to main content

Canonical public API

The QuorumKit API lives under include/quorumkit/. It is the surface new applications should build against.

The headers​

include/quorumkit/
  types.h           shared types: PeerId, GroupId, error codes
  node.h            creating and operating a Raft node
  admin.h           add/remove peer, transfer leader, reset peers
  discovery.h       peer discovery and route table
  storage.h         storage backend contracts
  util.h            small utilities
  extensions/
    filesystem.h    filesystem snapshot adaptor
    throttle.h      snapshot throttle

node.h is the center. It defines the node lifecycle: construct a node, configure its peers, start it, apply operations through it, snapshot it, shut it down. The other headers support that lifecycle -- types.h gives the shared vocabulary, admin.h and discovery.h cover cluster management, and storage.h defines the contracts that storage backends implement.

What the API avoids​

The old braft headers pull in brpc types, bthread types, protobuf service bases, and storage implementation details. The QuorumKit headers do not. You can read node.h and understand the node model without knowing anything about brpc or bthread.

This is a deliberate choice. Transport, threading, and storage are internal concerns. The public API talks about nodes, peers, groups, and operations -- the concepts you actually think about when you are writing a replicated service.

The extensions/ directory holds optional pieces (filesystem adaptors, throttle hooks) that not every user needs. They are public but separate, so they do not clutter the core headers.