kvrpcpb.proto

syntax = "proto3";
package kvrpcpb;

import "metapb.proto";
import "errorpb.proto";
import "gogoproto/gogo.proto";
import "rustproto.proto";
import "deadlock.proto";
import "tracepb.proto";
import "resource_manager.proto";

option (gogoproto.marshaler_all) = true;
option (gogoproto.sizer_all) = true;
option (gogoproto.unmarshaler_all) = true;
option (rustproto.lite_runtime_all) = true;

option java_package = "org.tikv.kvproto";

// This proto file defines requests, responses, and helper messages for KV and raw
// APIs of TiKV (see tikvpb.proto).

// Transactional commands.

// A transactional get command. Lookup a value for `key` in the transaction with
// starting timestamp = `version`.
message GetRequest {
    Context context = 1;
    bytes key = 2;
    uint64 version = 3;
}

message GetResponse {
    // A region error indicates that the request was sent to the wrong TiKV node
    // (or other, similar errors).
    errorpb.Error region_error = 1;
    // A value could not be retrieved due to the state of the database for the requested key.
    KeyError error = 2;
    // A successful result.
    bytes value = 3;
    // True if the key does not exist in the database.
    bool not_found = 4;
    reserved 5;
    // Time and scan details when processing the request.
    ExecDetailsV2 exec_details_v2 = 6;
}

// Scan fetches values for a range of keys; it is part of the transaction with
// starting timestamp = `version`.
message ScanRequest {
    Context context = 1;
    bytes start_key = 2;
    // The maximum number of results to return.
    uint32 limit = 3;
    uint64 version = 4;
    // Return only the keys found by scanning, not their values.
    bool key_only = 5;
    bool reverse = 6;
    // For compatibility, when scanning forward, the range to scan is [start_key, end_key), where start_key < end_key;
    // and when scanning backward, it scans [end_key, start_key) in descending order, where end_key < start_key.
    bytes end_key = 7;
    // If sample_step > 0, skips 'sample_step - 1' number of keys after each returned key.
    // locks are not checked.
    uint32 sample_step = 8;
}

message ScanResponse {
    errorpb.Error region_error = 1;
    // Each KvPair may contain a key error.
    repeated KvPair pairs = 2;
    // This KeyError exists when some key is locked but we cannot check locks of all keys.
    // In this case, `pairs` should be empty and the client should redo scanning all the keys
    // after resolving the lock.
    KeyError error = 3;
}

// A prewrite is the first phase of writing to TiKV. It contains all data to be written in a transaction.
// TiKV will write the data in a preliminary state. Data cannot be read until it has been committed.
// The client should only commit a transaction once all prewrites succeed.
message PrewriteRequest {
    // What kind of checks need to be performed for keys in a pessimistic transaction.
    enum PessimisticAction {
        // The key needn't be locked and no extra write conflict checks are needed.
        SKIP_PESSIMISTIC_CHECK = 0;
        // The key should have been locked at the time of prewrite.
        DO_PESSIMISTIC_CHECK = 1;
        // The key doesn't need a pessimistic lock. But we need to do data constraint checks.
        DO_CONSTRAINT_CHECK = 2;
    }

    // for_update_ts constriants that should be checked when prewriting a pessimistic transaction.
    message ForUpdateTSConstraint {
        // The index of key in the prewrite request that should be checked.
        uint32 index = 1;
        // The expected for_update_ts of the pessimistic lock of the key.
        uint64 expected_for_update_ts = 2;
    }

    Context context = 1;
    // The data to be written to the database.
    repeated Mutation mutations = 2;
    // The client picks one key to be primary (unrelated to the primary key concept in SQL). This
    // key's lock is the source of truth for the state of a transaction. All other locks due to a
    // transaction will point to the primary lock.
    bytes primary_lock = 3;
    // Identifies the transaction being written.
    uint64 start_version = 4;
    uint64 lock_ttl = 5;
    // TiKV can skip some checks, used for speeding up data migration.
    bool skip_constraint_check = 6;
    // For pessimistic transaction, some mutations don't need to be locked, for example, non-unique index key.
    // Keys with deferred constraint checks are not locked.
    repeated PessimisticAction pessimistic_actions = 7;
    // How many keys this transaction involves in this region.
    uint64 txn_size = 8;
    // For pessimistic transactions only; used to check if a conflict lock is already committed.
    uint64 for_update_ts = 9;
    // If min_commit_ts > 0, this is a large transaction request, the final commit_ts
    // will be inferred from `min_commit_ts`.
    uint64 min_commit_ts = 10;
    // When async commit is enabled, `secondaries` should be set as the key list of all secondary
    // locks if the request prewrites the primary lock.
    bool use_async_commit = 11;
    repeated bytes secondaries = 12;
    // When the transaction involves only one region, it's possible to commit the transaction
    // directly with 1PC protocol.
    bool try_one_pc = 13;
    // The max commit ts is reserved for limiting the commit ts of 1PC or async commit, which can be used to avoid
    // inconsistency with schema change.
    uint64 max_commit_ts = 14;
    // The level of assertion to use on this prewrte request.
    AssertionLevel assertion_level = 15;
    // for_update_ts constriants that should be checked when prewriting a pessimistic transaction.
    // See https://github.com/tikv/tikv/issues/14311
    repeated ForUpdateTSConstraint for_update_ts_constraints = 16;
}

message PrewriteResponse {
    errorpb.Error region_error = 1;
    repeated KeyError errors = 2;
    // 0 if the min_commit_ts is not ready or any other reason that async
    // commit cannot proceed. The client can then fallback to normal way to
    // continue committing the transaction if prewrite are all finished.
    uint64 min_commit_ts = 3;
    // When the transaction is successfully committed with 1PC protocol, this field will be set to
    // the commit ts of the transaction. Otherwise, if TiKV failed to commit it with 1PC or the
    // transaction is not 1PC, the value will be 0.
    uint64 one_pc_commit_ts = 4;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 5;
}

// Used to specify the behavior when a pessimistic lock request is woken up after waiting for another
// lock.
enum PessimisticLockWakeUpMode {
    // When woken up, returns WriteConflict error to the client and the client should retry if necessary.
    // In this mode, results of `return_values` or `check_existence` will be set to `values` and `not_founds`
    // fields of the PessimisticLockResponse, which is compatible with old versions.
    WakeUpModeNormal = 0;
    // When woken up, continue trying to lock the key. This implicitly enables the `allow_lock_with_conflict`
    // behavior, which means, allow acquiring the lock even if there is WriteConflict on the key.
    // In this mode, `return_values` or `check_existence` fields of PessimisticLockResponse won't be used, and
    // all results are carried in the `results` field.
    WakeUpModeForceLock = 1;
}

// Lock a set of keys to prepare to write to them.
message PessimisticLockRequest {
    Context context = 1;
    // In this case every `Op` of the mutations must be `PessimisticLock`.
    repeated Mutation mutations = 2;
    bytes primary_lock = 3;
    uint64 start_version = 4;
    uint64 lock_ttl = 5;
    // Each locking command in a pessimistic transaction has its own timestamp. If locking fails, then
    // the corresponding SQL statement can be retried with a later timestamp, TiDB does not need to
    // retry the whole transaction. The name comes from the `SELECT ... FOR UPDATE` SQL statement which
    // is a locking read. Each `SELECT ... FOR UPDATE` in a transaction will be assigned its own
    // timestamp.
    uint64 for_update_ts = 6;
    // If the request is the first lock request, we don't need to detect deadlock.
    bool is_first_lock = 7;
    // Time to wait for lock released in milliseconds when encountering locks.
    // 0 means using default timeout in TiKV. Negative means no wait.
    int64 wait_timeout = 8;
    // If it is true, TiKV will acquire the pessimistic lock regardless of write conflict
    // and return the latest value. It's only supported for single mutation.
    bool force = 9 [deprecated = true];
    // If it is true, TiKV will return values of the keys if no error, so TiDB can cache the values for
    // later read in the same transaction.
    // When 'force' is set to true, this field is ignored.
    bool return_values = 10;
    // If min_commit_ts > 0, this is large transaction proto, the final commit_ts
    // would be infered from min_commit_ts.
    uint64 min_commit_ts = 11;
    // If set to true, it means TiKV need to check if the key exists, and return the result in
    // the `not_founds` feild in the response. This works no matter if `return_values` is set. If
    // `return_values` is set, it simply makes no difference; otherwise, the `value` field of the
    // repsonse will be empty while the `not_founds` field still indicates the keys' existence.
    bool check_existence = 12;
    // TiKV lock the record only when it exists
    bool lock_only_if_exists = 13;
    // Specifies the behavior when the request is woken up after wating for lock of another transaction.
    PessimisticLockWakeUpMode wake_up_mode = 14;
}

enum PessimisticLockKeyResultType {
    LockResultNormal = 0;
    LockResultLockedWithConflict = 1;
    LockResultFailed = 2;
}

message PessimisticLockKeyResult {
    PessimisticLockKeyResultType type = 1;
    bytes value = 2;
    bool existence = 3;
    // We allow a key be locked when there is write conflict (latest commit_ts > for_update_ts).
    // In this case, the key is semantically locked by a newer for_update_ts.
    // For each requested key, the field is non-zero if the key is locked with write conflict, and it
    // equals to the commit_ts of the latest version of the specified key. The for_update_ts field
    // of the lock that's actually written to TiKV will also be this value. At the same time,
    // `value` and `existence` will be returned regardless to how `return_values` and
    // `check_existence` are set.
    uint64 locked_with_conflict_ts = 4;
    // Hint the client that resolving lock is not needed for this lock. For `PessimisticLock`
    // requests only.
    bool skip_resolving_lock = 11;
}

message PessimisticLockResponse {
    errorpb.Error region_error = 1;
    repeated KeyError errors = 2;
    // It carries the latest value and its commit ts if force in PessimisticLockRequest is true.
    uint64 commit_ts = 3 [deprecated = true];
    bytes value = 4 [deprecated = true];
    // The values is set if 'return_values' is true in the request and no error.
    // If 'force' is true, this field is not used.
    // Only used when `wake_up_mode` is `WakeUpModeNormal`.
    repeated bytes values = 5;
    // Indicates whether the values at the same index is correspond to an existing key.
    // In legacy TiKV, this field is not used even 'force' is false. In that case, an empty value indicates
    // two possible situations: (1) the key does not exist. (2) the key exists but the value is empty.
    // Only used when `wake_up_mode` is `WakeUpModeNormal`.
    repeated bool not_founds = 6;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 7;
    // Results of the request. Only used when `wake_up_mode` is `WakeUpModeForceLock`.
    repeated PessimisticLockKeyResult results = 8;
}

// Unlock keys locked using `PessimisticLockRequest`.
message PessimisticRollbackRequest {
    Context context = 1;
    uint64 start_version = 2;
    uint64 for_update_ts = 3;
    repeated bytes keys = 4;
}

message PessimisticRollbackResponse {
    errorpb.Error region_error = 1;
    repeated KeyError errors = 2;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 3;
}

// Used to update the lock_ttl of a psessimistic and/or large transaction to prevent it from been killed.
message TxnHeartBeatRequest {
    Context context = 1;
    // The key of the lock to update.
    bytes primary_lock = 2;
    // Start timestamp of the large transaction.
    uint64 start_version = 3;
    // The new TTL the sender would like.
    uint64 advise_lock_ttl = 4;
}

message TxnHeartBeatResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // The TTL actually set on the requested lock.
    uint64 lock_ttl = 3;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 4;
}

// CheckTxnStatusRequest checks the status of a transaction.
// If the transaction is rollbacked/committed, return that result.
// If the TTL of the transaction is exhausted, abort that transaction and inform the caller.
// Otherwise, returns the TTL information for the transaction.
// CheckTxnStatusRequest may also push forward the minCommitTS of a large transaction.
message CheckTxnStatusRequest {
    Context context = 1;
    // Primary key and lock ts together to locate the primary lock of a transaction.
    bytes primary_key = 2;
    // Starting timestamp of the transaction being checked.
    uint64 lock_ts = 3;
    // The start timestamp of the transaction which this request is part of.
    uint64 caller_start_ts = 4;
    // The client must specify the current time to TiKV using this timestamp. It is used to check TTL
    // timeouts. It may be inaccurate.
    uint64 current_ts = 5;
    // If true, then TiKV will leave a rollback tombstone in the write CF for `primary_key`, even if
    // that key is not locked.
    bool rollback_if_not_exist = 6;
    // This field is set to true only if the transaction is known to fall back from async commit.
    // Then, CheckTxnStatus treats the transaction as non-async-commit even if the use_async_commit
    // field in the primary lock is true.
    bool force_sync_commit = 7;
    // If the check request is used to resolve or decide the transaction status for a input pessimistic
    // lock, the transaction status could not be decided if the primary lock is pessimistic too and
    // it's still uncertain.
    bool resolving_pessimistic_lock = 8;
    // Whether it's needed to check if the lock on the key (if any) is the primary lock.
    // This is for handling some corner cases when a pessimistic transaction changes its primary
    // (see https://github.com/pingcap/tidb/issues/42937 for details). This field is necessary
    // because the old versions of clients cannot handle some results returned from TiKV correctly.
    // For new versions, this field should always be set to true.
    bool verify_is_primary = 9;
}

message CheckTxnStatusResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // Three kinds of transaction status:
    //   locked: lock_ttl > 0
    //   committed: commit_version > 0
    //   rollbacked: lock_ttl = 0 && commit_version = 0
    uint64 lock_ttl = 3;
    uint64 commit_version = 4;
    // The action performed by TiKV (and why if the action is to rollback).
    Action action = 5;
    LockInfo lock_info = 6;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 7;
}

// Part of the async commit protocol, checks for locks on all supplied keys. If a lock is missing,
// does not have a successful status, or belongs to another transaction, TiKV will leave a rollback
// tombstone for that key.
message CheckSecondaryLocksRequest {
    Context context = 1;
    repeated bytes keys = 2;
    // Identifies the transaction we are investigating.
    uint64 start_version = 3;
}

message CheckSecondaryLocksResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // For each key in `keys` in `CheckSecondaryLocks`, there will be a lock in
    // this list if there is a lock present and belonging to the correct transaction,
    // nil otherwise.
    repeated LockInfo locks = 3;
    // If any of the locks have been committed, this is the commit ts used. If no
    // locks have been committed, it will be zero.
    uint64 commit_ts = 4;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 5;
}

// The second phase of writing to TiKV. If there are no errors or conflicts, then this request
// commits a transaction so that its data can be read by other transactions.
message CommitRequest {
    reserved 5;
    reserved "binlog";
    Context context = 1;
    // Identifies the transaction.
    uint64 start_version = 2;
    // All keys in the transaction (to be committed).
    repeated bytes keys = 3;
    // Timestamp for the end of the transaction. Must be greater than `start_version`.
    uint64 commit_version = 4;
}

message CommitResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // If the commit ts is derived from min_commit_ts, this field should be set.
    uint64 commit_version = 3;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 4;
}

// Not yet implemented.
message ImportRequest {
    repeated Mutation mutations = 1;
    uint64 commit_version = 2;
}

message ImportResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

// Cleanup a key by possibly unlocking it.
// From 4.0 onwards, this message is no longer used.
message CleanupRequest {
    Context context = 1;
    bytes key = 2;
    uint64 start_version = 3;
    // The current timestamp, used in combination with a lock's TTL to determine
    // if the lock has expired. If `current_ts == 0`, then the key will be unlocked
    // irrespective of its TTL.
    uint64 current_ts = 4;
}

message CleanupResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // Set if the key is already committed.
    uint64 commit_version = 3;
}

// Similar to a `Get` request.
message BatchGetRequest {
    Context context = 1;
    repeated bytes keys = 2;
    uint64 version = 3;
}

message BatchGetResponse {
    errorpb.Error region_error = 1;
    repeated KvPair pairs = 2;
    reserved 3;
    // Time and scan details when processing the request.
    ExecDetailsV2 exec_details_v2 = 4;
    // This KeyError exists when some key is locked but we cannot check locks of all keys.
    // In this case, `pairs` should be empty and the client should redo batch get all the keys
    // after resolving the lock.
    KeyError error = 5;
}

// Rollback a prewritten transaction. This will remove the preliminary data from the database,
// unlock locks, and leave a rollback tombstone.
message BatchRollbackRequest {
    Context context = 1;
    // Identify the transaction to be rolled back.
    uint64 start_version = 2;
    // The keys to rollback.
    repeated bytes keys = 3;
}

message BatchRollbackResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 3;
}

// Scan the database for locks. Used at the start of the GC process to find all
// old locks.
message ScanLockRequest {
    Context context = 1;
    // Returns all locks with a start timestamp before `max_version`.
    uint64 max_version = 2;
    // Start scanning from this key.
    bytes start_key = 3;
    // The maximum number of locks to return.
    uint32 limit = 4;
    // The exclusive upperbound for scanning.
    bytes end_key = 5;
}

message ScanLockResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // Info on all locks found by the scan.
    repeated LockInfo locks = 3;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 4;
}

// For all keys locked by the transaction identified by `start_version`, either
// commit or rollback the transaction and unlock the key.
message ResolveLockRequest {
    Context context = 1;
    uint64 start_version = 2;
    // `commit_version == 0` means the transaction was rolled back.
    // `commit_version > 0` means the transaction was committed at the given timestamp.
    uint64 commit_version = 3;
    repeated TxnInfo txn_infos = 4;
    // Only resolve specified keys.
    repeated bytes keys = 5;
}

message ResolveLockResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
    // Execution details about the request processing.
    ExecDetailsV2 exec_details_v2 = 3;
}

// Request TiKV to garbage collect all non-current data older than `safe_point`.
message GCRequest {
    Context context = 1;
    uint64 safe_point = 2;
}

message GCResponse {
    errorpb.Error region_error = 1;
    KeyError error = 2;
}

// Delete a range of data from TiKV.
message DeleteRangeRequest {
    Context context = 1;
    bytes start_key = 2;
    bytes end_key = 3;
    // If true, the data will not be immediately deleted, but the operation will
    // still be replicated via Raft. This is used to notify TiKV that the data
    // will be deleted using `unsafe_destroy_range` soon.
    bool notify_only = 4;
}

message DeleteRangeResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

// Preparing the flashback for a region/key range will "lock" the region
// so that there is no any read, write or schedule operation could be proposed before
// the actual flashback operation.
message PrepareFlashbackToVersionRequest {
    Context context = 1;
    bytes start_key = 2;
    bytes end_key = 3;
    // The `start_ts` which we will use to write a lock to prevent
    // the `resolved_ts` from advancing during the whole process.
    uint64 start_ts = 4;
    // The TS version which the data will flashback to later.
    uint64 version = 5;
}

message PrepareFlashbackToVersionResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

// Flashback the region to a specific point with the given `version`, please
// make sure the region is "locked" by `PrepareFlashbackToVersionRequest` first,
// otherwise this request will fail.
message FlashbackToVersionRequest {
    Context context = 1;
    // The TS version which the data should flashback to.
    uint64 version = 2;
    bytes start_key = 3;
    bytes end_key = 4;
    // The `start_ts` and `commit_ts` which the newly written MVCC version will use.
    // Please make sure the `start_ts` is the same one in `PrepareFlashbackToVersionRequest`.
    uint64 start_ts = 5;
    uint64 commit_ts = 6;
}

message FlashbackToVersionResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

// Raw commands.

message RawGetRequest {
    Context context = 1;
    bytes key = 2;
    string cf = 3;
}

message RawGetResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    bytes value = 3;
    bool not_found = 4;
}

message RawBatchGetRequest {
    Context context = 1;
    repeated bytes keys = 2;
    string cf = 3;
}

message RawBatchGetResponse {
    errorpb.Error region_error = 1;
    repeated KvPair pairs = 2;
}

message RawPutRequest {
    Context context = 1;
    bytes key = 2;
    bytes value = 3;
    string cf = 4;
    uint64 ttl = 5;
    bool for_cas = 6;
}

message RawPutResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

message RawBatchPutRequest {
    Context context = 1;
    repeated KvPair pairs = 2;
    string cf = 3;
    uint64 ttl = 4 [deprecated=true];
    bool for_cas = 5;

    // The time-to-live for each keys in seconds, and if the length of `ttls`
    // is exactly one, the ttl will be applied to all keys. Otherwise, the length
    // mismatch between `ttls` and `pairs` will return an error.
    repeated uint64 ttls = 6;
}

message RawBatchPutResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

message RawDeleteRequest {
    Context context = 1;
    bytes key = 2;
    string cf = 3;
    bool for_cas = 4;
}

message RawDeleteResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

message RawBatchDeleteRequest {
    Context context = 1;
    repeated bytes keys = 2;
    string cf = 3;
    bool for_cas = 4;
}

message RawBatchDeleteResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

message RawScanRequest {
    Context context = 1;
    bytes start_key = 2;
    uint32 limit = 3;
    bool key_only = 4;
    string cf = 5;
    bool reverse = 6;
    // For compatibility, when scanning forward, the range to scan is [start_key, end_key), where start_key < end_key;
    // and when scanning backward, it scans [end_key, start_key) in descending order, where end_key < start_key.
    bytes end_key = 7;
}

message RawScanResponse {
    errorpb.Error region_error = 1;
    repeated KvPair kvs = 2;
}

message RawDeleteRangeRequest {
    Context context = 1;
    bytes start_key = 2;
    bytes end_key = 3;
    string cf = 4;
}

message RawDeleteRangeResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

message RawBatchScanRequest {
    Context context = 1;
    repeated KeyRange ranges = 2; // scanning range
    uint32 each_limit = 3; // max number of returning kv pairs for each scanning range
    bool key_only = 4;
    string cf = 5;
    bool reverse = 6;
}

message RawBatchScanResponse {
    errorpb.Error region_error = 1;
    repeated KvPair kvs = 2;
}

// Store commands (sent to a whole TiKV cluster, rather than a certain region).

message UnsafeDestroyRangeRequest {
    Context context = 1;
    bytes start_key = 2;
    bytes end_key = 3;
}

message UnsafeDestroyRangeResponse {
    errorpb.Error region_error = 1;
    string error = 2;
}

message RegisterLockObserverRequest {
    Context context = 1;
    uint64 max_ts = 2;
}

message RegisterLockObserverResponse {
    string error = 1;
}

message CheckLockObserverRequest {
    Context context = 1;
    uint64 max_ts = 2;
}

message CheckLockObserverResponse {
    string error = 1;
    bool is_clean = 2;
    repeated LockInfo locks = 3;
}

message RemoveLockObserverRequest {
    Context context = 1;
    uint64 max_ts = 2;
}

message RemoveLockObserverResponse {
    string error = 1;
}

message PhysicalScanLockRequest {
    Context context = 1;
    uint64 max_ts = 2;
    bytes start_key = 3;
    uint32 limit = 4;
}

message PhysicalScanLockResponse {
    string error = 1;
    repeated LockInfo locks = 2;
}

// Sent from PD to a TiKV node.
message SplitRegionRequest {
    Context context = 1;
    bytes split_key = 2 [deprecated=true];
    repeated bytes split_keys = 3; // when use it to do batch split, `split_key` should be empty.
    // Once enabled, the split_key will not be encoded.
    bool is_raw_kv = 4;
}

message SplitRegionResponse {
    errorpb.Error region_error = 1;
    metapb.Region left  = 2 [deprecated=true]; // set when there are only 2 result regions.
    metapb.Region right = 3 [deprecated=true]; // set when there are only 2 result regions.
    repeated metapb.Region regions = 4; // include all result regions.
}

// Sent from TiFlash to a TiKV node.
message ReadIndexRequest{
    Context context = 1;

    // TiKV checks the given range if there is any unapplied lock
    // blocking the read request.
    uint64 start_ts = 2;
    repeated KeyRange ranges = 3;
}

message ReadIndexResponse{
    errorpb.Error region_error = 1;
    uint64 read_index = 2;
    // If `locked` is set, this read request is blocked by a lock.
    // The lock should be returned to the client.
    kvrpcpb.LockInfo locked = 3;
}

// Commands for debugging transactions.

message MvccGetByKeyRequest {
    Context context = 1;
    bytes key = 2;
}

message MvccGetByKeyResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    MvccInfo info = 3;
}

message MvccGetByStartTsRequest {
    Context context = 1;
    uint64 start_ts = 2;
}

message MvccGetByStartTsResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    bytes key = 3;
    MvccInfo info = 4;
}

// Helper messages.

// Miscellaneous metadata attached to most requests.
message Context {
    reserved 4;
    reserved "read_quorum";
    uint64 region_id = 1;
    metapb.RegionEpoch region_epoch = 2;
    metapb.Peer peer = 3;
    uint64 term = 5;
    CommandPri priority = 6;
    IsolationLevel isolation_level = 7;
    bool not_fill_cache = 8;
    bool sync_log = 9;

    // True means execution time statistics should be recorded and returned.
    bool record_time_stat = 10;
    // True means RocksDB scan statistics should be recorded and returned.
    bool record_scan_stat = 11;

    bool replica_read = 12;
    // Read requests can ignore locks belonging to these transactions because either
    // these transactions are rolled back or theirs commit_ts > read request's start_ts.
    repeated uint64 resolved_locks = 13;
    uint64 max_execution_duration_ms = 14;

    // After a region applies to `applied_index`, we can get a
    // snapshot for the region even if the peer is a follower.
    uint64 applied_index = 15;
    // A hint for TiKV to schedule tasks more fairly. Query with same task ID
    // may share same priority and resource quota.
    uint64 task_id = 16;

    // Not required to read the most up-to-date data, replicas with `safe_ts` >= `start_ts`
    // can handle read request directly
    bool stale_read = 17;

    // Any additional serialized information about the request.
    bytes resource_group_tag = 18;

    // Used to tell TiKV whether operations are allowed or not on different disk usages.
    DiskFullOpt disk_full_opt = 19;

    // Indicates the request is a retry request and the same request may have been sent before.
    bool is_retry_request = 20;

    // API version implies the encode of the key and value.
    APIVersion api_version = 21;

    // Read request should read through locks belonging to these transactions because these
    // transactions are committed and theirs commit_ts <= read request's start_ts.
    repeated uint64 committed_locks = 22;

    // The informantion to trace a request sent to TiKV.
    tracepb.TraceContext trace_context = 23;

    // The source of the request, will be used as the tag of the metrics reporting.
    // This field can be set for any requests that require to report metrics with any extra labels.
    string request_source = 24;
    
    // The source of the current transaction.
    uint64 txn_source = 25;

    reserved 26; // Used for resource_group_name, now it's moved into ResourceControlContext.

    // If `busy_threshold_ms` is given, TiKV can reject the request and return a `ServerIsBusy`
    // error before processing if the estimated waiting duration exceeds the threshold.
    uint32 busy_threshold_ms = 27;

    // Some information used for resource control.
    ResourceControlContext resource_control_context = 28;

    // The keyspace that the request is sent to.
    // NOTE: This field is only meaningful while the api_version is V2.
    uint32 keyspace_id = 32;
    
    // The buckets version that the request is sent to.
    // NOTE: This field is only meaningful while enable buckets.
    uint64 buckets_version = 33;
    
    // It tells us where the request comes from in TiDB. If it isn't from TiDB, leave it blank.
    // This is for tests only and thus can be safely changed/removed without affecting compatibility.
    SourceStmt source_stmt = 34;
}

message ResourceControlContext {
    // It's used to identify which resource group the request belongs to.
    string resource_group_name = 1;

    // The resource consumption of the resource group that have completed at all TiKVs between the previous request to this TiKV and current request.
    // It's used as penalty to make the local resource scheduling on one TiKV takes the gloabl resource consumption into consideration.
    resource_manager.Consumption penalty = 2;

    // This priority would override the original priority of the resource group for the request.
    // Used to deprioritize the runaway queries. 
    uint64 override_priority = 3;
}

// The API version the server and the client is using.
// See more details in https://github.com/tikv/rfcs/blob/master/text/0069-api-v2.md.
enum APIVersion {
    // `V1` is mainly for TiDB & TxnKV, and is not safe to use RawKV along with the others.
    // V1 server only accepts V1 requests. V1 raw requests with TTL will be rejected.
    V1 = 0;
    //
    // `V1TTL` is only available to RawKV, and 8 bytes representing the unix timestamp in
    // seconds for expiring time will be append to the value of all RawKV entries. For example:
    // ------------------------------------------------------------
    // | User value     | Expire Ts                               |
    // ------------------------------------------------------------
    // | 0x12 0x34 0x56 | 0x00 0x00 0x00 0x00 0x00 0x00 0xff 0xff |
    // ------------------------------------------------------------
    // V1TTL server only accepts V1 raw requests.
    // V1 client should not use `V1TTL` in request. V1 client should always send `V1`.
    V1TTL = 1;
    //
    // `V2` use new encoding for RawKV & TxnKV to support more features.
    //
    // Key Encoding:
    //  TiDB: start with `m` or `t`, the same as `V1`.
    //  TxnKV: prefix with `x`, encoded as `MCE( x{keyspace id} + {user key} ) + timestamp`.
    //  RawKV: prefix with `r`, encoded as `MCE( r{keyspace id} + {user key} ) + timestamp`.
    //  Where the `{keyspace id}` is fixed-length of 3 bytes in network byte order.
    //  Besides, RawKV entires must be in `default` CF.
    //
    // Value Encoding:
    //  TiDB & TxnKV: the same as `V1`.
    //  RawKV: `{user value} + {optional fields} + {meta flag}`. The last byte in the
    //  raw value must be meta flags. For example:
    //  --------------------------------------
    //  | User value     | Meta flags        |
    //  --------------------------------------
    //  | 0x12 0x34 0x56 | 0x00 (0b00000000) |
    //  --------------------------------------
    //  Bit 0 of meta flags is for TTL. If set, the value contains 8 bytes expiring time as
    //  unix timestamp in seconds at the very left to the meta flags.
    //  --------------------------------------------------------------------------------
    //  | User value     | Expiring time                           | Meta flags        |
    //  --------------------------------------------------------------------------------
    //  | 0x12 0x34 0x56 | 0x00 0x00 0x00 0x00 0x00 0x00 0xff 0xff | 0x01 (0b00000001) |
    //  --------------------------------------------------------------------------------
    //  Bit 1 is for deletion. If set, the entry is logical deleted.
    //  ---------------------
    //  | Meta flags        |
    //  ---------------------
    //  | 0x02 (0b00000010) |
    //  ---------------------
    //
    // V2 server accpets V2 requests and V1 transactional requests that statrts with TiDB key
    // prefix (`m` and `t`).
    V2 = 2;
}

message SourceStmt {
    uint64 start_ts = 1;
    uint64 connection_id = 2;
    uint64 stmt_id = 3;
    // session alias set by user
    string session_alias = 4;
}

message LockInfo {
    bytes primary_lock = 1;
    uint64 lock_version = 2;
    bytes key = 3;
    uint64 lock_ttl = 4;
    // How many keys this transaction involves in this region.
    uint64 txn_size = 5;
    Op lock_type = 6;
    uint64 lock_for_update_ts = 7;
    // Fields for transactions that are using Async Commit.
    bool use_async_commit = 8;
    uint64 min_commit_ts = 9;
    repeated bytes secondaries = 10;
    // The time elapsed since last update of lock wait info when waiting.
    // It's used in timeout errors. 0 means unknown or not applicable. 
    // It can be used to help the client decide whether to try resolving the lock.
    uint64 duration_to_last_update_ms = 11;
}

message KeyError {
    LockInfo locked = 1; // Client should backoff or cleanup the lock then retry.
    string retryable = 2; // Client may restart the txn. e.g write conflict.
    string abort = 3; // Client should abort the txn.
    WriteConflict conflict = 4; // Write conflict is moved from retryable to here.
    AlreadyExist already_exist = 5; // Key already exists
    Deadlock deadlock = 6; // Deadlock is used in pessimistic transaction for single statement rollback.
    CommitTsExpired commit_ts_expired = 7; // Commit ts is earlier than min commit ts of a transaction.
    TxnNotFound txn_not_found = 8; // Txn not found when checking txn status.
    CommitTsTooLarge commit_ts_too_large = 9; // Calculated commit TS exceeds the limit given by the user.
    AssertionFailed assertion_failed = 10; // Assertion of a `Mutation` is evaluated as a failure.
    PrimaryMismatch primary_mismatch = 11; // CheckTxnStatus is sent to a lock that's not the primary.
}

message WriteConflict {
    enum Reason {
        Unknown = 0;
        Optimistic = 1; // in optimistic transactions.
        PessimisticRetry = 2; // a lock acquisition request waits for a lock and awakes, or meets a newer version of data, let TiDB retry.
        SelfRolledBack = 3; // the transaction itself has been rolled back when it tries to prewrite.
        RcCheckTs = 4; // RcCheckTs failure by meeting a newer version, let TiDB retry.
        LazyUniquenessCheck = 5; // write conflict found in lazy uniqueness check in pessimistic transactions.
    }

    uint64 start_ts = 1;
    uint64 conflict_ts = 2;
    bytes key = 3;
    bytes primary = 4;
    uint64 conflict_commit_ts = 5;
    Reason reason = 6;
}

message AlreadyExist {
    bytes key = 1;
}

message Deadlock {
    uint64 lock_ts = 1;
    bytes lock_key = 2;
    uint64 deadlock_key_hash = 3;
    repeated deadlock.WaitForEntry wait_chain = 4;
}

message CommitTsExpired {
    uint64 start_ts = 1;
    uint64 attempted_commit_ts = 2;
    bytes key = 3;
    uint64 min_commit_ts = 4;
}

message TxnNotFound {
    uint64 start_ts = 1;
    bytes primary_key = 2;
}

message CommitTsTooLarge {
    uint64 commit_ts = 1; // The calculated commit TS.
}

message AssertionFailed {
    uint64 start_ts = 1;
    bytes key = 2;
    Assertion assertion = 3;
    uint64 existing_start_ts = 4;
    uint64 existing_commit_ts = 5;
}

message PrimaryMismatch {
    LockInfo lock_info = 1;
}

enum CommandPri {
    Normal = 0; // Normal is the default value.
    Low = 1;
    High = 2;
}

enum IsolationLevel {
    SI = 0; // SI = snapshot isolation
    RC = 1; // RC = read committed
    RCCheckTS = 2; // RC read and it's needed to check if there exists more recent versions.
}

// Operation allowed info during each TiKV storage threshold.
enum DiskFullOpt {
    NotAllowedOnFull = 0; // The default value, means operations are not allowed either under almost full or already full.
    AllowedOnAlmostFull = 1; // Means operations will be allowed when disk is almost full.
    AllowedOnAlreadyFull = 2; // Means operations will be allowed when disk is already full.
}

message TimeDetail {
    // Off-cpu wall time elapsed in TiKV side. Usually this includes queue waiting time and
    // other kind of waitings in series. (Wait time in the raftstore is not included.)
    uint64 wait_wall_time_ms = 1;
    // Off-cpu and on-cpu wall time elapsed to actually process the request payload. It does not
    // include `wait_wall_time`.
    // This field is very close to the CPU time in most cases. Some wait time spend in RocksDB
    // cannot be excluded for now, like Mutex wait time, which is included in this field, so that
    // this field is called wall time instead of CPU time.
    uint64 process_wall_time_ms = 2;
    // KV read wall Time means the time used in key/value scan and get.
    uint64 kv_read_wall_time_ms = 3;
    // Total wall clock time spent on this RPC in TiKV .
    uint64 total_rpc_wall_time_ns = 4;
}

message TimeDetailV2 {
    // Off-cpu wall time elapsed in TiKV side. Usually this includes queue waiting time and
    // other kind of waitings in series. (Wait time in the raftstore is not included.)
    uint64 wait_wall_time_ns = 1;
    // Off-cpu and on-cpu wall time elapsed to actually process the request payload. It does not
    // include `wait_wall_time` and `suspend_wall_time`.
    // This field is very close to the CPU time in most cases. Some wait time spend in RocksDB
    // cannot be excluded for now, like Mutex wait time, which is included in this field, so that
    // this field is called wall time instead of CPU time.
    uint64 process_wall_time_ns = 2;
    // Cpu wall time elapsed that task is waiting in queue.
    uint64 process_suspend_wall_time_ns = 3;
    // KV read wall Time means the time used in key/value scan and get.
    uint64 kv_read_wall_time_ns = 4;
    // Total wall clock time spent on this RPC in TiKV .
    uint64 total_rpc_wall_time_ns = 5;
}

message ScanInfo {
    int64 total = 1;
    int64 processed = 2;
    int64 read_bytes = 3;
}

// Only reserved for compatibility.
message ScanDetail {
    ScanInfo write = 1;
    ScanInfo lock = 2;
    ScanInfo data = 3;
}

message ScanDetailV2 {
    // Number of user keys scanned from the storage.
    // It does not include deleted version or RocksDB tombstone keys.
    // For Coprocessor requests, it includes keys that has been filtered out by
    // Selection.
    uint64 processed_versions = 1;

    // Number of bytes of user key-value pairs scanned from the storage, i.e.
    // total size of data returned from MVCC layer.
    uint64 processed_versions_size = 8;

    // Approximate number of MVCC keys meet during scanning. It includes
    // deleted versions, but does not include RocksDB tombstone keys.
    //
    // When this field is notably larger than `processed_versions`, it means
    // there are a lot of deleted MVCC keys.
    uint64 total_versions = 2;

    // Total number of deletes and single deletes skipped over during
    // iteration, i.e. how many RocksDB tombstones are skipped.
    uint64 rocksdb_delete_skipped_count = 3;

    // Total number of internal keys skipped over during iteration.
    // See https://github.com/facebook/rocksdb/blob/9f1c84ca471d8b1ad7be9f3eebfc2c7e07dfd7a7/include/rocksdb/perf_context.h#L84 for details.
    uint64 rocksdb_key_skipped_count = 4;

    // Total number of RocksDB block cache hits.
    uint64 rocksdb_block_cache_hit_count = 5;

    // Total number of block reads (with IO).
    uint64 rocksdb_block_read_count = 6;

    // Total number of bytes from block reads.
    uint64 rocksdb_block_read_byte = 7;

    // Total time used for block reads.
    uint64 rocksdb_block_read_nanos = 9;

    // Time used for getting a raftstore snapshot (including proposing read index, leader confirmation and getting the RocksDB snapshot).
    uint64 get_snapshot_nanos = 10;

    // Time used for proposing read index from read pool to store pool, equals 0 when performing lease read.
    uint64 read_index_propose_wait_nanos = 11;

    // Time used for leader confirmation, equals 0 when performing lease read.
    uint64 read_index_confirm_wait_nanos = 12;

    // Time used for read pool scheduling.
    uint64 read_pool_schedule_wait_nanos = 13;
}

message ExecDetails {
    // Available when ctx.record_time_stat = true or meet slow query.
    TimeDetail time_detail = 1;

    // Available when ctx.record_scan_stat = true or meet slow query.
    ScanDetail scan_detail = 2;

    // See https://github.com/pingcap/kvproto/pull/689
    reserved 3;
    reserved 4;
}

message ExecDetailsV2 {
    // Available when ctx.record_time_stat = true or meet slow query.
    // deprecated. Should use `time_detail_v2` instead.
    TimeDetail time_detail = 1;

    // Available when ctx.record_scan_stat = true or meet slow query.
    ScanDetailV2 scan_detail_v2 = 2;

    // Raftstore writing durations of the request. Only available for some write requests.
    WriteDetail write_detail = 3;
    // Available when ctx.record_time_stat = true or meet slow query.
    TimeDetailV2 time_detail_v2 = 4;
}

message WriteDetail {
    // Wait duration in the store loop.
    uint64 store_batch_wait_nanos = 1;

    // Wait duration before sending proposal to peers.
    uint64 propose_send_wait_nanos = 2;

    // Total time spent on persisting the log.
    uint64 persist_log_nanos = 3;

    // Wait time until the Raft log write leader begins to write.
    uint64 raft_db_write_leader_wait_nanos = 4;

    // Time spent on synchronizing the Raft log to the disk.
    uint64 raft_db_sync_log_nanos = 5;

    // Time spent on writing the Raft log to the Raft memtable.
    uint64 raft_db_write_memtable_nanos = 6;

    // Time waiting for peers to confirm the proposal (counting from the instant when the leader sends the proposal message).
    uint64 commit_log_nanos = 7;

    // Wait duration in the apply loop.
    uint64 apply_batch_wait_nanos = 8;

    // Total time spend to applying the log.
    uint64 apply_log_nanos = 9;

    // Wait time until the KV RocksDB lock is acquired.
    uint64 apply_mutex_lock_nanos = 10;

    // Wait time until becoming the KV RocksDB write leader.
    uint64 apply_write_leader_wait_nanos = 11;

    // Time spent on writing the KV DB WAL to the disk.
    uint64 apply_write_wal_nanos = 12;

    // Time spent on writing to the memtable of the KV RocksDB.
    uint64 apply_write_memtable_nanos = 13;

    // Time spent on waiting in the latch.
    uint64 latch_wait_nanos = 14;

    // Processing time in the transaction layer.
    uint64 process_nanos = 15;

    // Wait time because of the scheduler flow control or quota limiter throttling.
    uint64 throttle_nanos = 16;

    // Wait time in the waiter manager for pessimistic locking.
    uint64 pessimistic_lock_wait_nanos = 17;
}

message KvPair {
    KeyError error = 1;
    bytes key = 2;
    bytes value = 3;
}

enum Op {
    Put = 0;
    Del = 1;
    Lock = 2;
    Rollback = 3;
    // insert operation has a constraint that key should not exist before.
    Insert = 4;
    PessimisticLock = 5;
    CheckNotExists = 6;
}

enum Assertion {
     None = 0;
     Exist = 1;
     NotExist = 2;
}

enum AssertionLevel {
    // No assertion.
    Off = 0;
    // Assertion is enabled, but not enforced when it might affect performance.
    Fast = 1;
    // Assertion is enabled and enforced.
    Strict = 2;
}

message Mutation {
    Op op = 1;
    bytes key = 2;
    bytes value = 3;
    Assertion assertion = 4;
}

message MvccWrite {
    Op type = 1;
    uint64 start_ts = 2;
    uint64 commit_ts = 3;
    bytes short_value = 4;
    bool has_overlapped_rollback = 5;
    bool has_gc_fence = 6;
    uint64 gc_fence = 7;
    uint64 last_change_ts = 8;
    uint64 versions_to_last_change = 9;
}

message MvccValue {
    uint64 start_ts = 1;
    bytes value = 2;
}

message MvccLock {
    Op type = 1;
    uint64 start_ts = 2;
    bytes primary = 3;
    bytes short_value = 4;
    uint64 ttl = 5;
    uint64 for_update_ts = 6;
    uint64 txn_size = 7;
    bool use_async_commit = 8;
    repeated bytes secondaries = 9;
    repeated uint64 rollback_ts = 10;
    uint64 last_change_ts = 11;
    uint64 versions_to_last_change = 12;
}

message MvccInfo {
    MvccLock lock = 1;
    repeated MvccWrite writes = 2;
    repeated MvccValue values = 3;
}

message TxnInfo {
    uint64 txn = 1;
    uint64 status = 2;
}

enum Action {
    NoAction = 0;
    TTLExpireRollback = 1;
    LockNotExistRollback = 2;
    MinCommitTSPushed = 3;
    TTLExpirePessimisticRollback = 4;
    LockNotExistDoNothing = 5;
}

message KeyRange {
    bytes start_key = 1;
    bytes end_key = 2;
}

enum ExtraOp {
    Noop = 0;
    // ReadOldValue represents to output the previous value for delete/update operations.
    ReadOldValue = 1;
}

message LeaderInfo {
    uint64 region_id = 1;
    uint64 peer_id = 2;
    uint64 term = 3;
    metapb.RegionEpoch region_epoch = 4;
    ReadState read_state = 5;
}

message ReadState {
    uint64 applied_index = 1;
    uint64 safe_ts = 2;
}

message CheckLeaderRequest {
    repeated LeaderInfo regions = 1;
    uint64 ts = 2;
}

message CheckLeaderResponse {
    repeated uint64 regions = 1;
    uint64 ts = 2;
}

message StoreSafeTSRequest {
    // Get the minimal `safe_ts` from regions that overlap with the key range [`start_key`, `end_key`)
    // An empty key range means all regions in the store
    KeyRange key_range = 1;
}

message StoreSafeTSResponse {
    uint64 safe_ts = 1;
}

message RawGetKeyTTLRequest {
    Context context = 1;
    bytes key = 2;
    string cf = 3;
}

message RawGetKeyTTLResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    uint64 ttl = 3;
    bool not_found = 4;
}

message RawCASRequest {
    Context context = 1;
    bytes key = 2;
    bytes value = 3;
    bool previous_not_exist = 4;
    bytes previous_value = 5;
    string cf = 6;
    uint64 ttl = 7;
}

message RawCASResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    bool succeed = 3;
    // The previous value regardless of whether the comparison is succeed.
    bool previous_not_exist = 4;
    bytes previous_value = 5;
}

message GetLockWaitInfoRequest {
    Context context = 1;
    // TODO: There may need some filter options to be used on conditional querying, e.g., finding
    // the lock waiting status for some specified transaction.
}

message GetLockWaitInfoResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    repeated deadlock.WaitForEntry entries = 3;
}

message GetLockWaitHistoryRequest {
    Context context = 1;
    // TODO: There may need some filter options to be used on conditional querying, e.g., finding
    // the lock waiting status for some specified transaction.
}

message GetLockWaitHistoryResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    repeated deadlock.WaitForEntry entries = 3;
}

message RawCoprocessorRequest {
    kvrpcpb.Context context = 1;

    string copr_name = 2;
    // Coprorcessor version constraint following SEMVER definition.
    string copr_version_req = 3;

    repeated KeyRange ranges = 4;
    bytes data = 5;
}

message RawCoprocessorResponse {
    errorpb.Error region_error = 1;
    // Error message for cases like if no coprocessor with a matching name is found
    // or on a version mismatch between plugin_api and the coprocessor.
    string error = 2;
    bytes data = 3;
}

enum ChecksumAlgorithm {
    Crc64_Xor = 0;
}

message RawChecksumRequest {
    Context context = 1;
    ChecksumAlgorithm algorithm = 2;
    repeated KeyRange ranges = 3;
}

message RawChecksumResponse {
    errorpb.Error region_error = 1;
    string error = 2;
    uint64 checksum = 3;
    uint64 total_kvs = 4;
    uint64 total_bytes = 5;
}

message CompactError {
    oneof error {
        CompactErrorInvalidStartKey err_invalid_start_key = 1;
        CompactErrorPhysicalTableNotExist err_physical_table_not_exist = 2;
        CompactErrorCompactInProgress err_compact_in_progress = 3;
        CompactErrorTooManyPendingTasks err_too_many_pending_tasks = 4;
    }
}

message CompactErrorInvalidStartKey {}

message CompactErrorPhysicalTableNotExist {}

message CompactErrorCompactInProgress {}

message CompactErrorTooManyPendingTasks {}

message CompactRequest {
    // If specified, the compaction will start from this start key.
    // If unspecified, the compaction will start from beginning.
    // NOTE 1: The start key should be never manually constructed. You should always use a key
    // returned in CompactResponse.
    // NOTE 2: the compaction range will be always restricted by physical_table_id.
    bytes start_key = 1;

    // The physical table that will be compacted.
    //
    // TODO: this is information that TiKV doesn't need to know.
    // See https://github.com/pingcap/kvproto/issues/912
    int64 physical_table_id = 2;

    // The logical table id of the compaction. When receiving parallel requests with the same
    // logical table id, err_compact_in_progress will be returned.
    //
    // TODO: this is information that TiKV doesn't need to know.
    // See https://github.com/pingcap/kvproto/issues/912
    int64 logical_table_id = 3;

    // API version of the request
    APIVersion api_version = 7;
    // Keyspace of the table located in.
    uint32 keyspace_id = 8;
}

message CompactResponse {
    CompactError error = 1;

    // The compaction is done incrementally. If there are more data to compact, this field
    // will be set. The client can request to compact more data according to the `compacted_end_key`.
    bool has_remaining = 2;

    bytes compacted_start_key = 3;
    bytes compacted_end_key = 4;
}

message TiFlashSystemTableRequest {
    string sql = 1;
}

message TiFlashSystemTableResponse {
    bytes data = 1 [(gogoproto.customtype) = "github.com/pingcap/kvproto/pkg/sharedbytes.SharedBytes", (gogoproto.nullable) = false];
}