client-rust/proto/raft_serverpb.proto

syntax = "proto3";
package raft_serverpb;

import "eraftpb.proto";
import "metapb.proto";
import "kvrpcpb.proto";
import "disk_usage.proto";
import "encryptionpb.proto";
import "rustproto.proto";

option (rustproto.lite_runtime_all) = true;

option java_package = "org.tikv.kvproto";

message RaftMessage {
    uint64 region_id = 1;
    metapb.Peer from_peer = 2;
    metapb.Peer to_peer = 3;
    eraftpb.Message message = 4;
    metapb.RegionEpoch region_epoch = 5;
    // true means to_peer is a tombstone peer and it should remove itself.
    bool is_tombstone = 6;
    // Region key range [start_key, end_key).
    bytes start_key = 7;
    bytes end_key = 8;
    // If it has value, to_peer should be removed if merge is never going to complete.
    metapb.Region merge_target = 9;
    ExtraMessage extra_msg = 10;
    bytes extra_ctx = 11;

    disk_usage.DiskUsage disk_usage = 12;
}

message RaftTruncatedState {
    uint64 index = 1;
    uint64 term = 2;
}

message SnapshotCFFile {
    string cf = 1;
    uint64 size = 2;
    uint32 checksum = 3;
}

message SnapshotMeta {
    repeated SnapshotCFFile cf_files = 1;
    // true means this snapshot is triggered for load balance
    bool for_balance = 2;
    // true means this is an empty snapshot for witness
    bool for_witness = 3;
    // the timestamp second to generate snapshot
    uint64 start = 4;
    // the duration of generating snapshot
    uint64 generate_duration_sec = 5;
    // the path of the tablet snapshot, it should only be used for v1 to receive
    // snapshot from v2
    string tablet_snap_path = 6;
    // A hint of the latest commit index on leader when sending snapshot.
    // It should only be used for v2 to send snapshot to v1.
    // See https://github.com/pingcap/tiflash/issues/7568
    uint64 commit_index_hint = 7;
}

message SnapshotChunk {
    RaftMessage message = 1;
    bytes data = 2;
}

message Done {}

message TabletSnapshotFileMeta {
    uint64 file_size = 1;
    string file_name = 2;
    // Some block data. Unencrypted.
    bytes head_chunk = 3;
    // trailing data including checksum. Unencrypted.
    bytes trailing_chunk = 4;
}

// Snapshot preview for server to decide whether skip some files.
// Server should send back an `AcceptedSnapshotFile` to let client
// keep sending specified files. Only SST files can be skipped, all
// other files should always be sent.
message TabletSnapshotPreview {
    repeated TabletSnapshotFileMeta metas = 1;
    // There may be too many metas, use a flag to indicate all metas
    // are sent.
    bool end = 2;
}

message TabletSnapshotFileChunk {
    uint64 file_size = 1;
    string file_name = 2;
    // Encrypted.
    bytes data = 3;
    // Initial vector if encryption is enabled.
    bytes iv = 4;
    encryptionpb.DataKey key = 5;
}

message TabletSnapshotHead {
    RaftMessage message = 1;
    bool use_cache = 2;
}

message TabletSnapshotEnd {
    // Checksum of all data sent in `TabletSnapshotFileChunk.data` and
    // `TabletSnapshotFileChunk.file_name`.
    uint64 checksum = 1;
}

message TabletSnapshotRequest {
    oneof payload {
        TabletSnapshotHead head = 1;
        TabletSnapshotPreview preview = 2;
        TabletSnapshotFileChunk chunk = 3;
        TabletSnapshotEnd end = 4;
    }
}

message AcceptedSnapshotFiles {
    repeated string file_name = 1;
}

message TabletSnapshotResponse {
    AcceptedSnapshotFiles files = 1;
}

message KeyValue {
    bytes key      = 1;
    bytes value    = 2;
}

message RaftSnapshotData {
    metapb.Region region = 1;
    uint64 file_size = 2;
    repeated KeyValue data = 3;
    uint64 version = 4;
    SnapshotMeta meta = 5;
    repeated metapb.Peer removed_records = 6;
    repeated MergedRecord merged_records = 7;
}

message StoreIdent {
    uint64 cluster_id = 1;
    uint64 store_id = 2;
    kvrpcpb.APIVersion api_version = 3;
}

message StoreRecoverState {
    // Used for TiKV start recovery when WAL of KVDB was disabled.
    // TiKV may read all relations between seqno and raft log index, and replay
    // all raft logs which corresponding seqno smaller than the seqno here.
    // After TiKV replays all raft logs and flushed KV data, the seqno here must
    // be updated.
    uint64 seqno = 1;
}

message RaftLocalState {
    eraftpb.HardState hard_state = 1;
    uint64 last_index = 2;
}

message RaftApplyState {
    uint64 applied_index = 1;
    uint64 last_commit_index = 3;
    uint64 commit_index = 4;
    uint64 commit_term = 5;
    RaftTruncatedState truncated_state = 2;
}

enum PeerState {
    Normal = 0;
    Applying = 1;
    Tombstone = 2;
    Merging = 3;
    // Currently used for witness to non-witness conversion: When a witness
    // has just become a non-witness, we need to set and persist this state,
    // so that when the service restarts before applying snapshot, we can
    // actively request snapshot when initializing this peer.
    Unavailable = 4;
}

message MergeState {
    uint64 min_index = 1;
    metapb.Region target = 2;
    uint64 commit = 3;
}

message MergedRecord {
    uint64 source_region_id = 1;
    metapb.RegionEpoch source_epoch = 2;
    // Peers of source region when merge is committed.
    repeated metapb.Peer source_peers = 3;
    // Removed peers (by confchange) of source region when merge is committed.
    repeated metapb.Peer source_removed_records = 9;
    uint64 target_region_id = 4;
    metapb.RegionEpoch target_epoch = 5;
    repeated metapb.Peer target_peers = 6;
    // Commit merge index.
    uint64 index = 7;
    // Prepare merge index.
    uint64 source_index = 8;
}

message RegionLocalState {
    PeerState state = 1;
    metapb.Region region = 2;
    MergeState merge_state = 3;
    // The apply index corresponding to the storage when it's initialized.
    uint64 tablet_index = 4;
    // Raft doesn't guarantee peer will be removed in the end. In v1, peer finds
    // out its destiny by logs or broadcast; in v2, leader is responsible to
    // ensure removed peers are destroyed.
    // Note: only peers who has been part of this region can be in this list.
    repeated metapb.Peer removed_records = 5;
    // Merged peer can't be deleted like gc peers. Instead, leader needs to
    // query target peer to decide whether source peer can be destroyed.
    repeated MergedRecord merged_records = 6;
}

message RegionSequenceNumberRelation {
    uint64 region_id = 1;
    uint64 sequence_number = 2;
    RaftApplyState apply_state = 3;
    RegionLocalState region_state = 4;
}

message AvailabilityContext {
    uint64 from_region_id = 1;
    metapb.RegionEpoch from_region_epoch = 2;
    bool unavailable = 3;
    bool trimmed = 4;
}

enum ExtraMessageType {
    MsgRegionWakeUp = 0;
    MsgWantRollbackMerge = 1;
    MsgCheckStalePeer = 2;
    MsgCheckStalePeerResponse = 3;
    // If leader is going to sleep, it will send requests to all its followers
    // to make sure they all agree to sleep.
    MsgHibernateRequest = 4;
    MsgHibernateResponse = 5;
    MsgRejectRaftLogCausedByMemoryUsage = 6;
    MsgAvailabilityRequest = 7;
    MsgAvailabilityResponse = 8;
    MsgVoterReplicatedIndexRequest = 9;
    MsgVoterReplicatedIndexResponse = 10;
    // Message means that `from` is tombstone. Leader can then update removed_records.
    MsgGcPeerRequest = 11;
    MsgGcPeerResponse = 12;
    MsgFlushMemtable = 13;
    MsgRefreshBuckets = 14;
}

message FlushMemtable {
    uint64 region_id = 1;
}

message RefreshBuckets {
    uint64 version = 1;
    repeated bytes keys = 2;
    repeated uint64 sizes = 3;
}

message CheckGcPeer {
    // The region ID who triggers the check and wait for report. It should be
    // the ID of RaftMessage.from.
    uint64 from_region_id = 1;
    // The region ID to be checked if should be destroyed.
    uint64 check_region_id = 2;
    // The epoch of the region to be checked.
    metapb.RegionEpoch check_region_epoch = 3;
    // The peer to be checked.
    metapb.Peer check_peer = 4;
}

message ExtraMessage {
    ExtraMessageType type = 1;
    // It's merge related index. In `WantRollbackMerge`, it's prepare merge index. In
    // `MsgGcPeerRequest`, it's the commit merge index. In `MsgVoterReplicatedIndexRequest`
    // it's the voter_replicated_index.
    uint64 index = 2;
    // In `MsgCheckStalePeerResponse`, it's the peers that receiver can continue to query.
    repeated metapb.Peer check_peers = 3;
    bool wait_data = 4;
    // Flag for forcely wake up hibernate regions if true.
    bool forcely_awaken = 5;
    CheckGcPeer check_gc_peer = 6;
    FlushMemtable flush_memtable = 7;
    // Used by `MsgAvailabilityRequest` and `MsgAvailabilityResponse` in v2.
    AvailabilityContext availability_context = 8;
    // notice the peer to refresh buckets version
    RefreshBuckets refresh_buckets = 9;
}