Cluster Management

WeftOS clustering extends the kernel to multi-node operation with peer tracking, failure detection, distributed process management, service advertisement, and data replication.

Source: crates/clawft-kernel/src/cluster.rs, mesh_process.rs, mesh_service_adv.rs, mesh_heartbeat.rs, mesh_chain.rs, mesh_tree.rs Feature: mesh (types), cluster (raft consensus)

NodeIdentity

Ed25519-based identity:

pub struct NodeIdentity { /* Ed25519 keypair */ }

impl NodeIdentity {
    pub fn generate() -> Self;
    pub fn sign(&self, data: &[u8]) -> Vec<u8>;
    pub fn verify(&self, data: &[u8], signature: &[u8]) -> bool;
    pub fn node_id(&self) -> String;  // hex(SHAKE-256(pubkey)[0..16])
}

NodeState Lifecycle

pub enum NodeState { Joining, Active, Suspect, Unreachable, Left }

SWIM Heartbeat

pub struct HeartbeatConfig {
    pub probe_interval_ms: u64,     // 1000
    pub ping_timeout_ms: u64,       // 500
    pub suspect_timeout_ms: u64,    // 5000
    pub indirect_witnesses: usize,  // 3
}

State: Alive -> Suspect -> Dead. Indirect pings provide confirmation.

Distributed Process Table

LWW CRDT: merge() compares last_updated timestamps (newer wins).

pub struct ProcessAdvertisement {
    pub global_pid: GlobalPid,
    pub agent_type: String,
    pub capabilities: Vec<String>,
    pub status: ProcessStatus,
    pub resource_summary: ResourceSummary,
}

Query: find_by_type(), find_by_capability(), least_loaded_node().

ConsistentHashRing

Distributes PID assignment: ConsistentHashRing { ring: BTreeMap<u64, String> }.

Service Advertisement

pub struct ClusterServiceRegistry {
    services: HashMap<String, Vec<ServiceAdvertisement>>,
}

LWW merge per (service_name, node_id). resolve_preferred() tries specific node, falls back to replicas.

Chain Replication

ChainSyncRequest with after_sequence for incremental pull. detect_chain_fork() returns InSync, LocalBehind, RemoteBehind, or Forked.

Tree Synchronization

Merkle-based: compare_tree_roots() returns sync action. MerkleProof with verify().

Metadata Consensus

Raft-style (behind cluster feature): ConsensusRole { Follower, Candidate, Leader }.

CRDT Gossip

CrdtGossipState holds DistributedProcessTable + ClusterServiceRegistry for eventual consistency.