Struct MultiRaft 

pub struct MultiRaft { /* private fields */ }

Multi-Raft coordinator managing multiple Raft groups on a single node.

This coordinator:

• Manages all Raft groups hosted on this node
• Batches heartbeats across groups sharing the same leader
• Routes incoming RPCs to the correct group
• Collects Ready output from all groups for the caller to execute

Implementations

impl MultiRaft

pub fn propose_conf_change( &mut self, group_id: u64, change: &ConfChange, ) -> Result<(u64, u64)>

Propose a configuration change to a Raft group.

The change is serialized into the group’s Raft log as a regular entry with a distinguishing prefix byte. It replicates through the normal AppendEntries path and is applied by every follower replica when the entry commits (see apply_conf_change).

Single-voter vs. multi-voter groups

Single-voter groups commit inside node.propose itself (see nodedb_raft::node::RaftNode::propose single-voter branch). In that case the commit has already happened by the time we return, so we safely apply the change inline: any caller that reads routing immediately after the propose sees the final state.

Multi-voter groups commit asynchronously once enough followers have replicated the entry. The apply then happens on the tick loop after it observes the updated commit_index. We MUST NOT inline-apply in that case — if the leader steps down before replication completes, a new leader may truncate the log entry and the local state would be permanently ahead of the committed state with no rollback path. Callers that need to wait for the apply should poll the routing table (see raft_loop::join::wait_for_routing_contains_learner).

Returns (group_id, log_index) on success.

pub fn apply_conf_change( &mut self, group_id: u64, change: &ConfChange, ) -> Result<()>

Apply a committed configuration change to this node’s view of the given Raft group.

This is called from the tick loop for every committed entry detected as a conf-change (via ConfChange::from_entry_data). It must be idempotent with respect to no-op changes so replaying the log after a crash does not double-apply.

impl MultiRaft

pub fn new(node_id: u64, routing: RoutingTable, data_dir: PathBuf) -> Self

pub fn with_election_timeout(self, min: Duration, max: Duration) -> Self

Configure election timeout range.

pub fn with_heartbeat_interval(self, interval: Duration) -> Self

Configure heartbeat interval.

pub fn add_group(&mut self, group_id: u64, peers: Vec<u64>) -> Result<()>

Initialize a Raft group on this node as a voting member.

peers is the list of other voters in the group (excluding self). For a learner-start group, use add_group_as_learner instead.

pub fn add_group_as_learner( &mut self, group_id: u64, voters: Vec<u64>, learners: Vec<u64>, ) -> Result<()>

Initialize a Raft group on this node as a non-voting learner.

The local node boots in the Learner role and will not stand for election until it is promoted by a PromoteLearner conf change.

voters is the full voter set of the group (excluding self). learners is the learner set of the group excluding self — usually empty unless multiple learners are being admitted in the same round.

pub fn tick(&mut self) -> MultiRaftReady

Tick all Raft groups. Returns aggregated ready output.

pub fn routing(&self) -> &RoutingTable

pub fn routing_mut(&mut self) -> &mut RoutingTable

pub fn node_id(&self) -> u64

pub fn group_count(&self) -> usize

pub fn groups_mut(&mut self) -> &mut HashMap<u64, RaftNode<RedbLogStorage>>

Mutable access to the underlying Raft groups (for testing / bootstrap).

pub fn group_statuses(&self) -> Vec<GroupStatus>

Snapshot of all Raft group states for observability.

pub fn leader_for_vshard(&self, vshard_id: u32) -> Result<Option<u64>>

Get the leader for a given vShard (from local group state).

pub fn propose(&mut self, vshard_id: u32, data: Vec<u8>) -> Result<(u64, u64)>

Propose a command to the Raft group that owns the given vShard.

Returns (group_id, log_index) on success.

pub fn is_group_leader(&self, group_id: u64) -> bool

Returns true if this node is currently the leader of group_id.

Returns false when the group does not exist on this node or when the node is a follower, candidate, or learner in the group.

pub fn propose_to_group(&mut self, group_id: u64, data: Vec<u8>) -> Result<u64>

Propose a command directly to a specific Raft group (e.g. the metadata group, which has no vShard mapping).

Returns the committed log index on success.

pub fn read_committed_entries( &self, group_id: u64, lo: u64, hi: u64, ) -> Result<Vec<LogEntry>>

Read committed log entries for a Raft group in the inclusive index range [lo, hi].

hi is clamped to the group’s commit_index so callers that pass u64::MAX never read uncommitted entries.

Used by the Calvin scheduler’s rebuild path to replay sequenced transactions from the sequencer Raft log after a restart.

Returns Err(ClusterError::Raft(RaftError::LogCompacted)) if lo has been compacted into a snapshot (caller must install a snapshot instead of replaying from log).

impl MultiRaft

pub fn commit_index_for(&self, group_id: u64) -> Option<u64>

Current commit index for a group, or None if the group is not hosted on this node.

pub fn ready_learners(&self, group_id: u64) -> Vec<u64>

Learners in group_id whose match_index on this leader has caught up to the current commit_index — safe to promote.

Returns an empty vec if this node is not the leader of the group or the group is not hosted here.

pub fn group_leader(&self, group_id: u64) -> u64

Observed leader id for a group (0 = unknown / no election yet).

pub fn group_role_is_leader(&self, group_id: u64) -> bool

Whether this node is currently the leader of group_id.

impl MultiRaft

pub fn handle_append_entries( &mut self, req: &AppendEntriesRequest, ) -> Result<AppendEntriesResponse>

Route an AppendEntries RPC to the correct group.

pub fn handle_request_vote( &mut self, req: &RequestVoteRequest, ) -> Result<RequestVoteResponse>

Route a RequestVote RPC to the correct group.

pub fn handle_install_snapshot( &mut self, req: &InstallSnapshotRequest, ) -> Result<InstallSnapshotResponse>

Route an InstallSnapshot RPC to the correct group.

pub fn snapshot_metadata(&self, group_id: u64) -> Result<(u64, u64, u64)>

Get the current term and snapshot metadata for a group (for building InstallSnapshot RPCs).

pub fn handle_append_entries_response( &mut self, group_id: u64, peer: u64, resp: &AppendEntriesResponse, ) -> Result<()>

Handle AppendEntries response for a specific group.

pub fn handle_request_vote_response( &mut self, group_id: u64, peer: u64, resp: &RequestVoteResponse, ) -> Result<()>

Handle RequestVote response for a specific group.

pub fn advance_applied(&mut self, group_id: u64, applied_to: u64) -> Result<()>

Advance applied index for a group after processing committed entries.

pub fn match_index_for(&self, group_id: u64, peer: u64) -> Option<u64>

Query a peer’s match_index from a specific Raft group’s leader state.

pub fn last_applied(&self, group_id: u64) -> Option<u64>

Read the locally-applied index for a Raft group hosted on this node. Returns None if the group is not mounted here.

Used by the tick loop to mirror last_applied into the per-group crate::applied_watcher::AppliedIndexWatcher — covers both the regular apply path and the snapshot-install path (which sets last_applied = last_included_index directly without producing committed entries).

pub fn applied_indices(&self) -> Vec<(u64, u64)>

(group_id, last_applied) pairs for every locally-mounted group. Cheap O(groups) snapshot — groups are few (one metadata + handful of vshard groups per node).