improve batch efficiency for high throughput workloads

config/client.go

@@ -48,6 +48,14 @@ const (
 	DefGrpcInitialWindowSize      = 1 << 27 // 128MiB
 	DefGrpcInitialConnWindowSize  = 1 << 27 // 128MiB
 	DefMaxConcurrencyRequestLimit = math.MaxInt64
+	DefBatchPolicy                = BatchPolicyStandard
+)
+
+const (
+	BatchPolicyBasic    = "basic"
+	BatchPolicyStandard = "standard"
+	BatchPolicyPositive = "positive"
+	BatchPolicyCustom   = "custom"
 )
 
 // TiKVClient is the config for tikv client.
@@ -72,6 +80,9 @@ type TiKVClient struct {
 	// CommitTimeout is the max time which command 'commit' will wait.
 	CommitTimeout string      `toml:"commit-timeout" json:"commit-timeout"`
 	AsyncCommit   AsyncCommit `toml:"async-commit" json:"async-commit"`
+
+	// BatchPolicy is the policy for batching requests.
+	BatchPolicy string `toml:"batch-policy" json:"batch-policy"`
 	// MaxBatchSize is the max batch size when calling batch commands API.
 	MaxBatchSize uint `toml:"max-batch-size" json:"max-batch-size"`
 	// If TiKV load is greater than this, TiDB will wait for a while to avoid little batch.
@@ -153,6 +164,7 @@ func DefaultTiKVClient() TiKVClient {
 			AllowedClockDrift: 500 * time.Millisecond,
 		},
 
+		BatchPolicy:       DefBatchPolicy,
 		MaxBatchSize:      128,
 		OverloadThreshold: 200,
 		MaxBatchWaitTime:  0,

internal/client/client.go

@@ -300,8 +300,7 @@ func (a *connArray) Init(addr string, security config.Security, idleNotify *uint
 	allowBatch := (cfg.TiKVClient.MaxBatchSize > 0) && enableBatch
 	if allowBatch {
 		a.batchConn = newBatchConn(uint(len(a.v)), cfg.TiKVClient.MaxBatchSize, idleNotify)
-		a.pendingRequests = metrics.TiKVBatchPendingRequests.WithLabelValues(a.target)
-		a.batchSize = metrics.TiKVBatchRequests.WithLabelValues(a.target)
+		a.batchConn.initMetrics(a.target)
 	}
 	keepAlive := cfg.TiKVClient.GrpcKeepAliveTime
 	keepAliveTimeout := cfg.TiKVClient.GrpcKeepAliveTimeout
@@ -365,6 +364,7 @@ func (a *connArray) Init(addr string, security config.Security, idleNotify *uint
 				dialTimeout:      a.dialTimeout,
 				tryLock:          tryLock{sync.NewCond(new(sync.Mutex)), false},
 				eventListener:    eventListener,
+				metrics:          &a.metrics,
 			}
 			batchClient.maxConcurrencyRequestLimit.Store(cfg.TiKVClient.MaxConcurrencyRequestLimit)
 			a.batchCommandsClients = append(a.batchCommandsClients, batchClient)