mocra-cluster

Embedded control plane for the mocra distributed crawler framework: a self-organizing Raft + redb cluster providing strongly-consistent leader election, distributed locks (with fencing tokens), KV/CAS, membership, and partition ownership — no external coordinator (ZooKeeper / etcd / Redis) required.

Features

• openraft 0.9 consensus over a redb replicated state machine + persistent log (crash-recoverable; snapshot / log compaction).
• HTTP + msgpack inter-node RPC; any node accepts writes (auto-forwarded to the leader), so "register to any node to join" works.
• Distributed locks with monotonic fencing tokens; compare-and-set; KV.
• Membership API + /cluster/join; dynamic scale up / down; leader failover; graceful shutdown.
• Partition ownership via rendezvous (HRW) hashing (≈1/N keys move on membership change) plus Raft-fenced ownership leases.

Example

use mocra_cluster::RaftControlPlane;
use std::time::Duration;

// bootstrap a single-node cluster (or start_cluster_node + init_cluster / join_cluster)
let cp = RaftControlPlane::start_single_node(1, "./data").await?;
cp.wait_leader(Duration::from_secs(10)).await?;

cp.set(b"key", b"value").await?;                         // strongly-consistent write via Raft
let token = cp.acquire_lock("job", "me", 5000).await?;  // fenced distributed lock
let owned = cp.owned_partitions();                      // partitions this node owns

Part of the mocra workspace.

License

Licensed under either of MIT or Apache-2.0 at your option.