importinto: calculate distsql scan concurency in nextgen (#63341) (#63362)

ref #61702

Author: ti-chi-bot

pkg/disttask/framework/scheduler/BUILD.bazel

@@ -26,6 +26,7 @@ go_library(
         "//pkg/kv",
         "//pkg/lightning/log",
         "//pkg/sessionctx",
+        "//pkg/sessionctx/vardef",
         "//pkg/util",
         "//pkg/util/backoff",
         "//pkg/util/cpu",
@@ -59,7 +60,7 @@ go_test(
     embed = [":scheduler"],
     flaky = True,
     race = "off",
-    shard_count = 38,
+    shard_count = 39,
     deps = [
         "//pkg/config",
         "//pkg/config/kerneltype",

pkg/disttask/framework/scheduler/autoscaler.go

@@ -23,6 +23,7 @@ import (
 	"github.com/pingcap/tidb/pkg/disttask/framework/handle"
 	"github.com/pingcap/tidb/pkg/disttask/framework/storage"
 	"github.com/pingcap/tidb/pkg/kv"
+	"github.com/pingcap/tidb/pkg/sessionctx/vardef"
 	"github.com/pingcap/tidb/pkg/util/cpu"
 	"github.com/pingcap/tidb/pkg/util/intest"
 	"github.com/pingcap/tidb/pkg/util/logutil"
@@ -39,7 +40,11 @@ const (
 	// To improve performance for small tasks, we assume that on an 8c machine,
 	// importing 200 GiB of data requires full utilization of a single node’s resources.
 	// Therefore, for every additional 25 GiB, add 1 concurrency unit as an estimate for task concurrency.
-	baseSizePerConc = 25 * units.GiB
+	baseSizePerConc                = 25 * units.GiB
+	maxNodeCountLimitForImportInto = 32
+	// this value is calculated by 256/8, we have test on a 8c machine with 256
+	// concurrency, it's fast enough for checksum. we can tune this later if needed.
+	maxDistSQLConcurrencyPerCore = 32
 )
 
 // CalcMaxNodeCountByTableSize calculates the maximum number of nodes to execute DXF based on the table size.
@@ -49,7 +54,7 @@ func CalcMaxNodeCountByTableSize(size int64, coresPerNode int) int {
 
 // CalcMaxNodeCountByDataSize calculates the maximum number of nodes to execute DXF based on the data size.
 func CalcMaxNodeCountByDataSize(size int64, coresPerNode int) int {
-	return calcMaxNodeCountBySize(size, coresPerNode, 32)
+	return calcMaxNodeCountBySize(size, coresPerNode, maxNodeCountLimitForImportInto)
 }
 
 func calcMaxNodeCountBySize(size int64, coresPerNode int, factor float64) int {
@@ -112,3 +117,20 @@ func GetExecCPUNode(ctx context.Context) (int, error) {
 	}
 	return cpuNode, nil
 }
+
+// CalcDistSQLConcurrency calculates the DistSQL concurrency based on the thread
+// count, max node count and CPU cores of each node.
+// when maxNodeCnt <= 1,we use task concurrency times DefDistSQLScanConcurrency,
+// else, we use a linear interpolation method to gradually increase the concurrency
+// to maxDistSQLConcurrencyPerCore*nodeCPU.
+func CalcDistSQLConcurrency(threadCnt, maxNodeCnt, nodeCPU int) int {
+	if maxNodeCnt <= 1 {
+		return threadCnt * vardef.DefDistSQLScanConcurrency
+	}
+
+	start := vardef.DefDistSQLScanConcurrency * nodeCPU
+	interval := nodeCPU * (maxDistSQLConcurrencyPerCore - vardef.DefDistSQLScanConcurrency)
+	totalStepCount := maxNodeCountLimitForImportInto - 1
+	stepCount := min(totalStepCount, maxNodeCnt-1)
+	return int(float64(start) + float64(interval)*float64(stepCount)/float64(totalStepCount))
+}

pkg/disttask/framework/scheduler/autoscaler_test.go

@@ -102,3 +102,42 @@ func TestCalcConcurrencyByDataSize(t *testing.T) {
 			fmt.Sprintf("dataSize:%d cores:%d", tt.dataSize, tt.cores))
 	}
 }
+
+func TestCalcDistSQLConcurrency(t *testing.T) {
+	tests := []struct {
+		// concurrency, nodeCount, nodeCoreCount
+		c, n, nc int
+		expected int
+	}{
+		// on 8c machine
+		{c: 1, n: 1, nc: 8, expected: 15},
+		{c: 3, n: 1, nc: 8, expected: 45},
+		{c: 7, n: 1, nc: 8, expected: 105},
+		{c: 8, n: 1, nc: 8, expected: 120},
+		{c: 8, n: 2, nc: 8, expected: 124},
+		{c: 8, n: 5, nc: 8, expected: 137},
+		{c: 8, n: 32, nc: 8, expected: 256},
+		{c: 8, n: 33, nc: 8, expected: 256},
+		{c: 8, n: 50, nc: 8, expected: 256},
+		// on 16c machine
+		{c: 1, n: 1, nc: 16, expected: 15},
+		{c: 7, n: 1, nc: 16, expected: 105},
+		{c: 16, n: 1, nc: 16, expected: 240},
+		{c: 16, n: 5, nc: 16, expected: 275},
+		{c: 16, n: 32, nc: 16, expected: 512},
+		{c: 16, n: 33, nc: 16, expected: 512},
+		{c: 16, n: 50, nc: 16, expected: 512},
+		// on 32c machine
+		{c: 1, n: 1, nc: 32, expected: 15},
+		{c: 7, n: 1, nc: 32, expected: 105},
+		{c: 32, n: 1, nc: 32, expected: 480},
+		{c: 32, n: 5, nc: 32, expected: 550},
+		{c: 32, n: 32, nc: 32, expected: 1024},
+		{c: 32, n: 33, nc: 32, expected: 1024},
+	}
+	for i, tt := range tests {
+		t.Run(fmt.Sprintf("case-%d", i), func(t *testing.T) {
+			require.Equal(t, tt.expected, CalcDistSQLConcurrency(tt.c, tt.n, tt.nc))
+		})
+	}
+}

pkg/executor/importer/import.go

@@ -1304,9 +1304,11 @@ func (e *LoadDataController) InitDataFiles(ctx context.Context) error {
 		failpoint.InjectCall("mockImportDataSize", &totalSize)
 		e.ThreadCnt = scheduler.CalcConcurrencyByDataSize(totalSize, targetNodeCPUCnt)
 		e.MaxNodeCnt = scheduler.CalcMaxNodeCountByDataSize(totalSize, targetNodeCPUCnt)
-		e.logger.Info("set import thread count for nextgen kernel",
+		e.DistSQLScanConcurrency = scheduler.CalcDistSQLConcurrency(e.ThreadCnt, e.MaxNodeCnt, targetNodeCPUCnt)
+		e.logger.Info("auto calculate resource related params",
 			zap.Int("thread count", e.ThreadCnt),
 			zap.Int("max node count", e.MaxNodeCnt),
+			zap.Int("dist sql scan concurrency", e.DistSQLScanConcurrency),
 			zap.Int("target node cpu count", targetNodeCPUCnt),
 			zap.String("total file size", units.BytesSize(float64(totalSize))))
 	}
@@ -1612,6 +1614,3 @@ func GetTargetNodeCPUCnt(ctx context.Context, sourceType DataSourceType, path st
 	}
 	return handle.GetCPUCountOfNode(ctx)
 }
-
-// TestSyncCh is used in unit test to synchronize the execution.
-var TestSyncCh = make(chan struct{})