Enum Command 

pub enum Command {
    Inbound(Inbound),
    Event(Event),
    FetchResponse {
        from: NodeId,
        body: Bytes,
    },
    VoteResponse {
        from: NodeId,
        body: Bytes,
    },
    FetchSnapshotResponse {
        from: NodeId,
        body: Bytes,
    },
    Timer(TimerTick),
    SubmitChange {
        records: Vec<MetadataRecord>,
        expected_image: Option<Arc<MetadataImage>>,
        reply: Sender<Result<SubmitChangeResult, RaftError>>,
    },
    SubmitDelegationTokenMutations {
        mutations: Vec<DelegationTokenMutation>,
        reply: Sender<Result<SubmitChangeResult, RaftError>>,
    },
    Reconfigure {
        change: VoterChange,
        reply: Sender<Result<ReconfigOutcome, RaftError>>,
    },
    TriggerSnapshot {
        reply: Sender<Result<(), RaftError>>,
    },
    QuorumStateSnapshot {
        reply: Sender<QuorumStateSnapshot>,
    },
    MetadataFetch {
        fetch_offset: i64,
        max_size: ByteSize,
        replica: Option<ReplicaKey>,
        reply: Sender<MetadataFetchSlice>,
    },
    Shutdown,
    Halt,
}

Everything that arrives on the engine’s mpsc and drives one turn of the loop.

Variants

Inbound(Inbound)

An inbound peer RPC with a oneshot to reply on.

Event(Event)

Injects a core Event directly. This is the test and driver entry point.

FetchResponse

A Fetch RESPONSE the follower received from the leader. A Fetch response carries log records. The follower must truncate, append, and apply those records BEFORE it feeds the ReceiveFetchResponse event to the core, so this response gets its own command.

Fields

from: NodeId

The leader that answered, which is the responder peer.

body: Bytes

The raw encoded wire::PeerResponse::Fetch body.

VoteResponse

A Vote RESPONSE a nominee received from a voter. Beside the grant it names the voter’s epoch, the leader it knows and that leader’s endpoint, which the engine acts on before the core counts the grant, as Kafka’s maybeHandleCommonResponse does.

Fields

from: NodeId

body: Bytes

FetchSnapshotResponse

A FetchSnapshot RESPONSE the follower received from the leader. It carries snapshot bytes that the follower reassembles before it resumes. This mirrors the dedicated command path of FetchResponse.

Fields

from: NodeId

body: Bytes

Timer(TimerTick)

An election, fetch, or heartbeat timer fired. The loop maps it to the right core event after it reads the liveness state. A fetch tick re-polls instead of electing, unless the leader has been missed enough times.

SubmitChange

Handle op: append and commit a metadata batch as the leader. It replies once the batch is committed and applied, or it replies with a rejection.

Fields

records: Vec<MetadataRecord>

expected_image: Option<Arc<MetadataImage>>

reply: Sender<Result<SubmitChangeResult, RaftError>>

SubmitDelegationTokenMutations

Handle op: compare and mutate delegation-token state in one engine turn.

Fields

mutations: Vec<DelegationTokenMutation>

reply: Sender<Result<SubmitChangeResult, RaftError>>

Reconfigure

Handle op: append a KIP-853 control batch and optionally wait for it to commit under the new voter set.

Fields

change: VoterChange

reply: Sender<Result<ReconfigOutcome, RaftError>>

TriggerSnapshot

Handle op: snapshot the current image to a checkpoint.

Fields

reply: Sender<Result<(), RaftError>>

QuorumStateSnapshot

Handle op: read a structured snapshot of consensus state for DescribeQuorum.

Fields

reply: Sender<QuorumStateSnapshot>

MetadataFetch

Handle op: read a committed __cluster_metadata slice for an observer’s API_KEY_METADATA_FETCH (1004), encoded as Kafka record batches.

Fields

fetch_offset: i64

max_size: ByteSize

replica: Option<ReplicaKey>

The observer that asked, when it is a replica. A leader tracks its progress under this key.

reply: Sender<MetadataFetchSlice>

Shutdown

Stop the loop gracefully: a leader resigns and the loop runs on, within a bound, until the quorum has elected a successor.

Halt

Stop the loop at once, handing nothing over, as a halted process does.

Implementations

impl Command

pub fn caller_gone(&self) -> bool

Whether the caller of a command that replies stopped waiting for the reply. A command that sends no reply never has a caller to lose.