tsoracle-openraft-toolkit

Reusable glue for building services on top of openraft.

What's in the box

• declare_raft_types_ext! — wraps the upstream RaftTypeConfig declaration with multi-leader-per-term, OneshotResponder, and a curated set of other defaults. Consumers supply only the slots that actually vary (Node, AppData, AppDataResponse, SnapshotData).
• RocksdbLogStore<C, K> — generic RaftLogStorage + RaftLogReader implementation backed by RocksDB. The K: KeySpace parameter chooses between Flat (single-group: one raft instance per process) and GroupPrefixed (multi-group: N raft instances sharing column families, keyed by group id). Passes openraft's bundled storage conformance suite.
• Lifecycle helpers: bootstrap (Fresh / Reopen / Join), change_membership, add_learner, and leadership_events — a deduped stream of role-class transitions derived from Raft::metrics().
• Wire codec: encode / decode helpers using the [version_byte | bincode(payload)] framing common to raft RPCs and storage records.

Feature flags

• rocksdb-log-store (default) — pulls in rocksdb and exposes RocksdbLogStore. Disable if you bring your own storage backend.
• test-fakes — exposes in-memory test fixtures for downstream conformance suites. Off by default.

Out of scope (today)

The toolkit deliberately stops short of shipping:

• A RaftStateMachine adapter. State machines' broadcast and responder shapes vary widely between consumers, and the abstraction isn't load-bearing yet.
• A RaftNetworkV2 implementation. Single-group services can roll their own with the wire codec; multi-group routing belongs in the host runtime that owns the group lifecycle.
• A snapshot-streaming transport. Filesystem-backed sidecar transports are opinionated enough that lifting one would constrain its callers.
• Multi-group host orchestration. Hosting N raft instances per process is a deployment-shaped concern, not a library concern.

These may move into the toolkit when a clear shared shape emerges.