raft.proto

// Copyright 2015 The Cockroach Authors.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.

syntax = "proto3";
package cockroach.kv.kvserver.kvserverpb;
option go_package = "kvserverpb";

import "roachpb/errors.proto";
import "roachpb/internal_raft.proto";
import "roachpb/metadata.proto";
import "kv/kvserver/liveness/livenesspb/liveness.proto";
import "kv/kvserver/kvserverpb/state.proto";
import "etcd/raft/v3/raftpb/raft.proto";
import "gogoproto/gogo.proto";
import "util/tracing/tracingpb/recorded_span.proto";

// RaftHeartbeat is a request that contains the barebones information for a
// raftpb.MsgHeartbeat raftpb.Message. RaftHeartbeats are coalesced and sent
// in a RaftMessageRequest, and reconstructed by the receiver into individual
// raftpb.Message protos.
message RaftHeartbeat {
  uint64 range_id = 1 [(gogoproto.customname) = "RangeID",
      (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.RangeID"];
  uint32 from_replica_id = 2 [(gogoproto.customname) = "FromReplicaID",
      (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.ReplicaID"];
  uint32 to_replica_id = 3 [(gogoproto.customname) = "ToReplicaID",
      (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.ReplicaID"];
  uint64 term = 4;
  uint64 commit = 5;
  bool quiesce = 6;
  reserved 7;
  repeated kv.kvserver.liveness.livenesspb.Liveness lagging_followers_on_quiesce = 8 [(gogoproto.nullable) = false];
  // This field helps migrate in the lagging_followers_on_quiesce field. For
  // messages sent by versions of Cockroach that do not know about the
  // lagging_followers_on_quiesce field (i.e. v20.1), we need to assume that all
  // replicas are lagging, not that none of them are.
  //
  // TODO(nvanbenschoten): Migration path:
  // v20.2: Add this field. Always set to true when quiesce == true.
  //        Consult field on receiver. Consider all replicas "lagging"
  //        if not set to true on liveness change.
  // v22.1: Keep sending. Stop consulting.
  // v22.2: Remove field.
  bool lagging_followers_on_quiesce_accurate = 10;
}

// RaftMessageRequest is the request used to send raft messages using our
// protobuf-based RPC codec. If a RaftMessageRequest has a non-empty number of
// heartbeats or heartbeat_resps, the contents of the message field is treated
// as a dummy message and discarded. A coalesced heartbeat request's replica
// descriptor's range ID must be zero.
message RaftMessageRequest {
  uint64 range_id = 1 [(gogoproto.customname) = "RangeID",
      (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.RangeID"];
  // Optionally, the start key of the sending replica. This is only populated
  // as a "hint" under certain conditions.
  bytes range_start_key = 8 [(gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.RKey"];

  roachpb.ReplicaDescriptor from_replica = 2 [(gogoproto.nullable) = false];
  roachpb.ReplicaDescriptor to_replica = 3 [(gogoproto.nullable) = false];

  raftpb.Message message = 4 [(gogoproto.nullable) = false];

  // Is this a quiesce request? A quiesce request is a MsgHeartbeat
  // which is requesting the recipient to stop ticking its local
  // replica as long as the current Raft state matches the heartbeat
  // Term/Commit. If the Term/Commit match, the recipient is marked as
  // quiescent. If they don't match, the message is passed along to
  // Raft which will generate a MsgHeartbeatResp that will unquiesce
  // the sender.
  bool quiesce = 5;
  // If quiesce is true, this set contains liveness information about
  // the replicas that were dead when the leader decided to quiesce and
  // were lagging behind the quiescence log index (meaning they would
  // have prevented quiescence had they been alive). If any replica
  // (leader or follower) becomes aware that a replica in this set has
  // become live, it should unquiesce the range so that the replica can
  // be caught back up.
  repeated kv.kvserver.liveness.livenesspb.Liveness lagging_followers_on_quiesce = 9 [(gogoproto.nullable) = false];

  // A coalesced heartbeat request is any RaftMessageRequest with a nonzero number of
  // heartbeats or heartbeat_resps.
  repeated RaftHeartbeat heartbeats = 6 [(gogoproto.nullable) = false];
  repeated RaftHeartbeat heartbeat_resps = 7 [(gogoproto.nullable) = false];

  reserved 10;
}

message RaftMessageRequestBatch {
  repeated RaftMessageRequest requests = 1 [(gogoproto.nullable) = false];
}

message RaftMessageResponseUnion {
  option (gogoproto.onlyone) = true;

  kv.kvpb.Error error = 1;
}

// RaftMessageResponse may be sent to the sender of a
// RaftMessageRequest. RaftMessage does not use the usual
// request/response pattern; it is primarily modeled as a one-way
// stream of requests. Normal 'responses' are usually sent as new
// requests on a separate stream in the other direction.
// RaftMessageResponse is not sent for every RaftMessageRequest, but
// may be used for certain error conditions.
message RaftMessageResponse {
  uint64 range_id = 1 [(gogoproto.customname) = "RangeID",
      (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.RangeID"];

  roachpb.ReplicaDescriptor from_replica = 2 [(gogoproto.nullable) = false];
  roachpb.ReplicaDescriptor to_replica = 3 [(gogoproto.nullable) = false];

  RaftMessageResponseUnion union = 4 [(gogoproto.nullable) = false];
}

// SnapshotRequest is the request used to send streaming snapshot requests.
message SnapshotRequest {
  enum Priority {
    UNKNOWN = 0;
    // RECOVERY is used for a Raft-initiated snapshots and for
    // up-replication snapshots (i.e. when a dead node has been
    // removed and the range needs to be up-replicated).
    RECOVERY = 1;
    // REBALANCE is used for snapshots involved in rebalancing.
    REBALANCE = 2;
  }

  enum Strategy {
    // KV_BATCH snapshots stream batches of KV pairs for all keys in a
    // range from the sender the the receiver. These KV pairs are then
    // combined into a large RocksDB WriteBatch that is atomically
    // applied.
    KV_BATCH = 0;
  }

  // Type is used for metrics collection on the receiver side. See
  // applySnapshot in replica_raftstorage.go.
  enum Type {
    // VIA_SNAPSHOT_QUEUE indicates the snapshots sent by the raft snapshot
    // queue to all types of replicas.
    VIA_SNAPSHOT_QUEUE = 0;
    // INITIAL indicates the initial snapshots sent to LEARNER (before they're
    // promoted to full voters) and NON_VOTER replicas for upreplication.
    //
    // As of the time of writing, we only send this snapshot from the
    // initializeRaftLearners after creating a new LEARNER or NON_VOTER replica.
    INITIAL = 1;
    reserved 2;
  }

  message Header {
    // The replica state at the time the snapshot was generated. Note
    // that ReplicaState.Desc differs from the above range_descriptor
    // field which holds the updated descriptor after the new replica
    // has been added while ReplicaState.Desc holds the descriptor
    // before the new replica has been added.
    storagepb.ReplicaState state = 5 [(gogoproto.nullable) = false];

    // The inner raft message is of type MsgSnap, and its snapshot data contains a UUID.
    RaftMessageRequest raft_message_request = 2 [(gogoproto.nullable) = false];

    // The estimated size of the range, to be used in reservation decisions.
    int64 range_size = 3;

    // The priority of the snapshot.
    Priority priority = 6;

    // The strategy of the snapshot.
    Strategy strategy = 7;

    // The type of the snapshot.
    Type type = 9;

    // Whether the snapshot uses the unreplicated RaftTruncatedState or not.
    // This is always true for snapshots generated in v21.1+ clusters. In v20.2
    // it was possible for ranges to be using the replicated variant. v21.1
    // therefore had code expecting that possibility (unlike v21.2 code, where
    // this field is assumed to always be true and thus never read). For
    // compatibility with v21.1 nodes however, v21.2 has to explicitly set this
    // field to true. In v22.1 we can remove it entirely seeing as how v21.2
    // code never reads the field.
    //
    // TODO(irfansharif): Remove this in v22.1.
    bool deprecated_unreplicated_truncated_state = 8;

    reserved 1, 4;
  }

  Header header = 1;

  // A RocksDB BatchRepr. Multiple kv_batches may be sent across multiple request messages.
  bytes kv_batch = 2 [(gogoproto.customname) = "KVBatch"];

  bool final = 4;

  reserved 3;
}

message SnapshotResponse {
  enum Status {
    UNKNOWN = 0;
    ACCEPTED = 1;
    APPLIED = 2;
    ERROR = 3;
    reserved 4;
  }
  Status status = 1;
  string message = 2;
  reserved 3;
}

// DelegateSnapshotRequest is the request used to delegate send snapshot requests.
message DelegateSnapshotRequest {
  uint64 range_id = 1 [(gogoproto.customname) = "RangeID",
    (gogoproto.casttype) = "github.com/cockroachdb/cockroach/pkg/roachpb.RangeID"];

  // The replica that delegates the snapshot request, in most cases the leader/leaseholder.
  // The snapshot request should originate from the coordinator.
  roachpb.ReplicaDescriptor coordinator_replica = 2 [(gogoproto.nullable) = false];

  // The replica receiving the snapshot.
  roachpb.ReplicaDescriptor recipient_replica = 3 [(gogoproto.nullable) = false];

  // The replica selected to act as the snapshot sender.
  roachpb.ReplicaDescriptor delegated_sender = 4 [(gogoproto.nullable) = false];

  // The priority of the snapshot.
  SnapshotRequest.Priority priority = 5;

  // The type of the snapshot.
  SnapshotRequest.Type type = 6;

  // The Raft applied term of the coordinator replica.
  // The term is used during snapshot receiving to reject messages from an older term.
  uint64 term = 7;

  // The truncated state of the Raft log on the coordinator replica.
  roachpb.RaftTruncatedState truncated_state = 8;
}

message DelegateSnapshotResponse {
  SnapshotResponse snapResponse = 1;
  // collected_spans stores trace spans recorded during the execution of this
  // request.
  repeated util.tracing.tracingpb.RecordedSpan collected_spans = 2 [(gogoproto.nullable) = false];
}

// ConfChangeContext is encoded in the raftpb.ConfChange.Context field.
message ConfChangeContext {
  string command_id = 1 [(gogoproto.customname) = "CommandID"];

  // Payload is the application-level command (i.e. an encoded
  // kvserverpb.RaftCommand).
  bytes payload = 2;
}