external engine: support concurrent read and split into smaller data

br/pkg/lightning/backend/backend.go

@@ -103,12 +103,13 @@ type LocalEngineConfig struct {
 
 // ExternalEngineConfig is the configuration used for local backend external engine.
 type ExternalEngineConfig struct {
-	StorageURI string
-	DataFiles  []string
-	StatFiles  []string
-	MinKey     []byte
-	MaxKey     []byte
-	SplitKeys  [][]byte
+	StorageURI      string
+	DataFiles       []string
+	StatFiles       []string
+	MinKey          []byte
+	MaxKey          []byte
+	SplitKeys       [][]byte
+	RegionSplitSize int64
 	// TotalFileSize can be an estimated value.
 	TotalFileSize int64
 	// TotalKVCount can be an estimated value.

br/pkg/lightning/backend/external/BUILD.bazel

@@ -23,6 +23,7 @@ go_library(
         "//br/pkg/lightning/backend/encode",
         "//br/pkg/lightning/backend/kv",
         "//br/pkg/lightning/common",
+        "//br/pkg/lightning/config",
         "//br/pkg/lightning/log",
         "//br/pkg/membuf",
         "//br/pkg/storage",

br/pkg/lightning/backend/external/engine.go

@@ -18,32 +18,34 @@ import (
 	"bytes"
 	"context"
 	"encoding/hex"
+	"slices"
 	"sort"
 	"time"
 
 	"github.com/cockroachdb/pebble"
 	"github.com/pingcap/errors"
 	"github.com/pingcap/tidb/br/pkg/lightning/common"
+	"github.com/pingcap/tidb/br/pkg/lightning/config"
 	"github.com/pingcap/tidb/br/pkg/lightning/log"
 	"github.com/pingcap/tidb/br/pkg/membuf"
 	"github.com/pingcap/tidb/br/pkg/storage"
 	"github.com/pingcap/tidb/kv"
 	"github.com/pingcap/tidb/util/logutil"
 	"go.uber.org/atomic"
 	"go.uber.org/zap"
+	"golang.org/x/sync/errgroup"
 )
 
 // Engine stored sorted key/value pairs in an external storage.
 type Engine struct {
-	storage    storage.ExternalStorage
-	dataFiles  []string
-	statsFiles []string
-	minKey     []byte
-	maxKey     []byte
-	splitKeys  [][]byte
-	bufPool    *membuf.Pool
-
-	iter *MergeKVIter
+	storage         storage.ExternalStorage
+	dataFiles       []string
+	statsFiles      []string
+	minKey          []byte
+	maxKey          []byte
+	splitKeys       [][]byte
+	regionSplitSize int64
+	bufPool         *membuf.Pool
 
 	keyAdapter         common.KeyAdapter
 	duplicateDetection bool
@@ -66,6 +68,7 @@ func NewExternalEngine(
 	minKey []byte,
 	maxKey []byte,
 	splitKeys [][]byte,
+	regionSplitSize int64,
 	keyAdapter common.KeyAdapter,
 	duplicateDetection bool,
 	duplicateDB *pebble.DB,
@@ -81,6 +84,7 @@ func NewExternalEngine(
 		minKey:             minKey,
 		maxKey:             maxKey,
 		splitKeys:          splitKeys,
+		regionSplitSize:    regionSplitSize,
 		bufPool:            membuf.NewPool(),
 		keyAdapter:         keyAdapter,
 		duplicateDetection: duplicateDetection,
@@ -97,9 +101,84 @@ func NewExternalEngine(
 // LoadIngestData loads the data from the external storage to memory in [start,
 // end) range, so local backend can ingest it. The used byte slice of ingest data
 // are allocated from Engine.bufPool and must be released by
-// MemoryIngestData.Finish(). For external.Engine, LoadIngestData must be called
-// with strictly increasing start / end key.
-func (e *Engine) LoadIngestData(ctx context.Context, start, end []byte) (common.IngestData, error) {
+// MemoryIngestData.DecRef().
+func (e *Engine) LoadIngestData(
+	ctx context.Context,
+	regionRanges []common.Range,
+	outCh chan<- common.DataAndRange,
+) error {
+	// estimate we will open at most 1000 files, so if e.dataFiles is small we can
+	// try to concurrently process ranges.
+	concurrency := int(MergeSortOverlapThreshold) / len(e.dataFiles)
+	numPerGroup := len(regionRanges) / concurrency
+	rangeGroups := make([][]common.Range, 0, concurrency)
+
+	if numPerGroup > 0 {
+		for i := 0; i < concurrency-1; i++ {
+			rangeGroups = append(rangeGroups, regionRanges[i*numPerGroup:(i+1)*numPerGroup])
+		}
+	}
+	rangeGroups = append(rangeGroups, regionRanges[(concurrency-1)*numPerGroup:])
+
+	eg, egCtx := errgroup.WithContext(ctx)
+	for _, ranges := range rangeGroups {
+		ranges := ranges
+		eg.Go(func() error {
+			iter, err := e.createMergeIter(egCtx, ranges[0].Start)
+			if err != nil {
+				return errors.Trace(err)
+			}
+			defer iter.Close()
+
+			if !iter.Next() {
+				return iter.Error()
+			}
+			for _, r := range ranges {
+				results, err := e.loadIngestData(egCtx, iter, r.Start, r.End)
+				if err != nil {
+					return errors.Trace(err)
+				}
+				for _, result := range results {
+					select {
+					case <-egCtx.Done():
+						return egCtx.Err()
+					case outCh <- result:
+					}
+				}
+			}
+			return nil
+		})
+	}
+	return eg.Wait()
+}
+
+func (e *Engine) buildIngestData(keys, values [][]byte, buf *membuf.Buffer) *MemoryIngestData {
+	return &MemoryIngestData{
+		keyAdapter:         e.keyAdapter,
+		duplicateDetection: e.duplicateDetection,
+		duplicateDB:        e.duplicateDB,
+		dupDetectOpt:       e.dupDetectOpt,
+		keys:               keys,
+		values:             values,
+		ts:                 e.ts,
+		memBuf:             buf,
+		refCnt:             atomic.NewInt64(0),
+		importedKVSize:     e.importedKVSize,
+		importedKVCount:    e.importedKVCount,
+	}
+}
+
+// LargeRegionSplitDataThreshold is exposed for test.
+var LargeRegionSplitDataThreshold = int(config.SplitRegionSize)
+
+// loadIngestData loads the data from the external storage to memory in [start,
+// end) range, and if the range is large enough, it will return multiple data.
+// The input `iter` should be called Next() before calling this function.
+func (e *Engine) loadIngestData(
+	ctx context.Context,
+	iter *MergeKVIter,
+	start, end []byte,
+) ([]common.DataAndRange, error) {
 	if bytes.Equal(start, end) {
 		return nil, errors.Errorf("start key and end key must not be the same: %s",
 			hex.EncodeToString(start))
@@ -109,54 +188,59 @@ func (e *Engine) LoadIngestData(ctx context.Context, start, end []byte) (common.
 	keys := make([][]byte, 0, 1024)
 	values := make([][]byte, 0, 1024)
 	memBuf := e.bufPool.NewBuffer()
-
-	if e.iter == nil {
-		iter, err := e.createMergeIter(ctx, start)
-		if err != nil {
-			return nil, errors.Trace(err)
-		}
-		e.iter = iter
-	} else {
-		// there should be a key that just exceeds the end key in last LoadIngestData
-		// invocation.
-		k, v := e.iter.Key(), e.iter.Value()
+	cnt := 0
+	size := 0
+	largeRegion := e.regionSplitSize > 2*int64(config.SplitRegionSize)
+	ret := make([]common.DataAndRange, 0, 1)
+	curStart := start
+
+	// there should be a key that just exceeds the end key in last loadIngestData
+	// invocation.
+	k, v := iter.Key(), iter.Value()
+	if len(k) > 0 {
 		keys = append(keys, memBuf.AddBytes(k))
 		values = append(values, memBuf.AddBytes(v))
+		cnt++
+		size += len(k) + len(v)
 	}
 
-	cnt := 0
-	for e.iter.Next() {
-		cnt++
-		k, v := e.iter.Key(), e.iter.Value()
+	for iter.Next() {
+		k, v = iter.Key(), iter.Value()
 		if bytes.Compare(k, start) < 0 {
 			continue
 		}
 		if bytes.Compare(k, end) >= 0 {
 			break
 		}
+		if largeRegion && size > LargeRegionSplitDataThreshold {
+			curKey := slices.Clone(k)
+			ret = append(ret, common.DataAndRange{
+				Data:  e.buildIngestData(keys, values, memBuf),
+				Range: common.Range{Start: curStart, End: curKey},
+			})
+			keys = make([][]byte, 0, 1024)
+			values = make([][]byte, 0, 1024)
+			size = 0
+			curStart = curKey
+		}
+
 		keys = append(keys, memBuf.AddBytes(k))
 		values = append(values, memBuf.AddBytes(v))
+		cnt++
+		size += len(k) + len(v)
 	}
-	if e.iter.Error() != nil {
-		return nil, errors.Trace(e.iter.Error())
+	if iter.Error() != nil {
+		return nil, errors.Trace(iter.Error())
 	}
 
 	logutil.Logger(ctx).Info("load data from external storage",
 		zap.Duration("cost time", time.Since(now)),
 		zap.Int("iterated count", cnt))
-	return &MemoryIngestData{
-		keyAdapter:         e.keyAdapter,
-		duplicateDetection: e.duplicateDetection,
-		duplicateDB:        e.duplicateDB,
-		dupDetectOpt:       e.dupDetectOpt,
-		keys:               keys,
-		values:             values,
-		ts:                 e.ts,
-		memBuf:             memBuf,
-		refCnt:             atomic.NewInt64(0),
-		importedKVSize:     e.importedKVSize,
-		importedKVCount:    e.importedKVCount,
-	}, nil
+	ret = append(ret, common.DataAndRange{
+		Data:  e.buildIngestData(keys, values, memBuf),
+		Range: common.Range{Start: curStart, End: end},
+	})
+	return ret, nil
 }
 
 func (e *Engine) createMergeIter(ctx context.Context, start kv.Key) (*MergeKVIter, error) {
@@ -227,13 +311,8 @@ func (e *Engine) SplitRanges(
 	return ranges, nil
 }
 
-// Close releases the resources of the engine.
-func (e *Engine) Close() error {
-	if e.iter == nil {
-		return nil
-	}
-	return errors.Trace(e.iter.Close())
-}
+// Close implements common.Engine.
+func (e *Engine) Close() error { return nil }
 
 // MemoryIngestData is the in-memory implementation of IngestData.
 type MemoryIngestData struct {

br/pkg/lightning/backend/external/split.go

@@ -121,6 +121,11 @@ func (r *RangeSplitter) Close() error {
 	return r.propIter.Close()
 }
 
+// GetRangeSplitSize returns the expected size of one range.
+func (r *RangeSplitter) GetRangeSplitSize() int64 {
+	return r.rangeSize
+}
+
 // SplitOneRangesGroup splits one group of ranges. `endKeyOfGroup` represents the
 // end key of the group, but it will be nil when the group is the last one.
 // `dataFiles` and `statFiles` are all the files that have overlapping key ranges

br/pkg/lightning/backend/local/engine.go

@@ -1046,8 +1046,19 @@ func (e *Engine) Finish(totalBytes, totalCount int64) {
 
 // LoadIngestData return (local) Engine itself because Engine has implemented
 // IngestData interface.
-func (e *Engine) LoadIngestData(_ context.Context, _, _ []byte) (common.IngestData, error) {
-	return e, nil
+func (e *Engine) LoadIngestData(
+	ctx context.Context,
+	regionRanges []common.Range,
+	outCh chan<- common.DataAndRange,
+) error {
+	for _, r := range regionRanges {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case outCh <- common.DataAndRange{Data: e, Range: r}:
+		}
+	}
+	return nil
 }
 
 type sstMeta struct {