chore(perf): Improve perf of bulk loader with Reuse allocator and ass…

…inging tags to allocator (#7360) (#7547)

* Reuse allocator (#7360)

Instead of creating a new z.Allocator for every encoder, this PR reuses the allocator. On a 21M dataset, bulk loader takes 2m22s on master, and only 1m35s on this PR. That's a major 35% performance improvement.

* use allocator tags

Co-authored-by: Manish R Jain <manish@dgraph.io>

codec/codec.go

@@ -47,7 +47,7 @@ type Encoder struct {
 	BlockSize int
 	pack      *pb.UidPack
 	uids      []uint64
-	alloc     *z.Allocator
+	Alloc     *z.Allocator
 	buf       *bytes.Buffer
 }
 
@@ -70,7 +70,7 @@ func (e *Encoder) packBlock() {
 	}
 
 	// Allocate blocks manually.
-	b := e.alloc.AllocateAligned(blockSize)
+	b := e.Alloc.AllocateAligned(blockSize)
 	block := (*pb.UidBlock)(unsafe.Pointer(&b[0]))
 
 	block.Base = e.uids[0]
@@ -106,7 +106,7 @@ func (e *Encoder) packBlock() {
 	}
 
 	sz := len(e.buf.Bytes())
-	block.Deltas = e.alloc.Allocate(sz)
+	block.Deltas = e.Alloc.Allocate(sz)
 	x.AssertTrue(sz == copy(block.Deltas, e.buf.Bytes()))
 	e.pack.Blocks = append(e.pack.Blocks, block)
 }
@@ -117,10 +117,12 @@ var tagEncoder string = "enc"
 func (e *Encoder) Add(uid uint64) {
 	if e.pack == nil {
 		e.pack = &pb.UidPack{BlockSize: uint32(e.BlockSize)}
-		e.alloc = z.NewAllocator(1024)
-		e.alloc.Tag = tagEncoder
 		e.buf = new(bytes.Buffer)
 	}
+	if e.Alloc == nil {
+		e.Alloc = z.NewAllocator(1024, "Encoder.Add")
+		e.Alloc.Tag = tagEncoder
+	}
 
 	size := len(e.uids)
 	if size > 0 && !match32MSB(e.uids[size-1], uid) {
@@ -138,8 +140,8 @@ func (e *Encoder) Add(uid uint64) {
 // Done returns the final output of the encoder. This UidPack MUST BE FREED via a call to FreePack.
 func (e *Encoder) Done() *pb.UidPack {
 	e.packBlock()
-	if e.pack != nil && e.alloc != nil {
-		e.pack.AllocRef = e.alloc.Ref
+	if e.pack != nil && e.Alloc != nil {
+		e.pack.AllocRef = e.Alloc.Ref
 	}
 	return e.pack
 }
@@ -405,7 +407,7 @@ func Decode(pack *pb.UidPack, seek uint64) []uint64 {
 // DecodeToBuffer is the same as Decode but it returns a z.Buffer which is
 // calloc'ed and can be SHOULD be freed up by calling buffer.Release().
 func DecodeToBuffer(pack *pb.UidPack, seek uint64) *z.Buffer {
-	buf, err := z.NewBufferWith(256<<20, 32<<30, z.UseCalloc)
+	buf, err := z.NewBufferWith(256<<20, 32<<30, z.UseCalloc, "Codec.DecodeToBuffer")
 	x.Check(err)
 	buf.AutoMmapAfter(1 << 30)
 

dgraph/cmd/bulk/count_index.go

@@ -132,10 +132,13 @@ func (c *countIndexer) writeIndex(buf *z.Buffer) {
 		fmt.Printf("Writing count index for %q rev=%v\n", pk.Attr, pk.IsReverse())
 	}
 
+	alloc := z.NewAllocator(8<<20, "CountIndexer.WriteIndex")
+	defer alloc.Release()
+
 	var pl pb.PostingList
-	encoder := codec.Encoder{BlockSize: 256}
+	encoder := codec.Encoder{BlockSize: 256, Alloc: alloc}
 
-	outBuf := z.NewBuffer(5 << 20)
+	outBuf := z.NewBuffer(5<<20, "CountIndexer.Buffer.WriteIndex")
 	defer outBuf.Release()
 	encode := func() {
 		pl.Pack = encoder.Done()
@@ -144,13 +147,13 @@ func (c *countIndexer) writeIndex(buf *z.Buffer) {
 		}
 
 		kv := posting.MarshalPostingList(&pl, nil)
-		codec.FreePack(pl.Pack)
 		kv.Key = append([]byte{}, lastCe.Key()...)
 		kv.Version = c.state.writeTs
 		kv.StreamId = streamId
 		badger.KVToBuffer(kv, outBuf)
 
-		encoder = codec.Encoder{BlockSize: 256}
+		alloc.Reset()
+		encoder = codec.Encoder{BlockSize: 256, Alloc: alloc}
 		pl.Reset()
 
 		// Flush out the buffer.

dgraph/cmd/bulk/mapper.go

@@ -59,7 +59,7 @@ type shardState struct {
 func newMapperBuffer(opt *options) *z.Buffer {
 	sz := float64(opt.MapBufSize) * 1.1
 	buf, err := z.NewBufferWithDir(int(sz), 2*int(opt.MapBufSize), z.UseMmap,
-		filepath.Join(opt.TmpDir, bufferDir))
+		filepath.Join(opt.TmpDir, bufferDir), "Mapper.Buffer")
 	x.Check(err)
 	return buf
 }

dgraph/cmd/bulk/reduce.go

@@ -356,7 +356,7 @@ func (r *reducer) startWriting(ci *countIndexer, writerCh chan *encodeRequest, c
 						StreamDone: true,
 					}
 
-					buf := z.NewBuffer(512)
+					buf := z.NewBuffer(512, "Reducer.Write")
 					defer buf.Release()
 					badger.KVToBuffer(doneKV, buf)
 
@@ -436,7 +436,8 @@ func bufferStats(cbuf *z.Buffer) {
 }
 
 func getBuf(dir string) *z.Buffer {
-	cbuf, err := z.NewBufferWithDir(64<<20, 64<<30, z.UseCalloc, filepath.Join(dir, bufferDir))
+	cbuf, err := z.NewBufferWithDir(64<<20, 64<<30, z.UseCalloc,
+		filepath.Join(dir, bufferDir), "Reducer.GetBuf")
 	x.Check(err)
 	cbuf.AutoMmapAfter(1 << 30)
 	return cbuf
@@ -549,7 +550,7 @@ func (r *reducer) toList(req *encodeRequest) {
 	pl := new(pb.PostingList)
 	writeVersionTs := r.state.writeTs
 
-	kvBuf := z.NewBuffer(260 << 20)
+	kvBuf := z.NewBuffer(260<<20, "Reducer.Buffer.ToList")
 	trackCountIndex := make(map[string]bool)
 
 	var freePostings []*pb.Posting
@@ -568,6 +569,12 @@ func (r *reducer) toList(req *encodeRequest) {
 		freePostings = append(freePostings, p)
 	}
 
+	alloc := z.NewAllocator(16<<20, "Reducer.ToList")
+	defer func() {
+		// We put alloc.Release in defer because we reassign alloc for split posting lists.
+		alloc.Release()
+	}()
+
 	start, end, num := cbuf.StartOffset(), cbuf.StartOffset(), 0
 	appendToList := func() {
 		if num == 0 {
@@ -594,7 +601,8 @@ func (r *reducer) toList(req *encodeRequest) {
 			}
 		}
 
-		enc := codec.Encoder{BlockSize: 256}
+		alloc.Reset()
+		enc := codec.Encoder{BlockSize: 256, Alloc: alloc}
 		var lastUid uint64
 		slice, next := []byte{}, start
 		for next >= 0 && (next < end || end == -1) {
@@ -628,7 +636,7 @@ func (r *reducer) toList(req *encodeRequest) {
 		// the full pb.Posting type is used (which pb.y contains the
 		// delta packed UID list).
 		if numUids == 0 {
-			codec.FreePack(pl.Pack)
+			// No need to FrePack here because we are reusing alloc.
 			return
 		}
 
@@ -656,6 +664,9 @@ func (r *reducer) toList(req *encodeRequest) {
 			kvs, err := l.Rollup(nil)
 			x.Check(err)
 
+			// Assign a new allocator, so we don't reset the one we were using during Rollup.
+			alloc = z.NewAllocator(16<<20, "Reducer.AppendToList")
+
 			for _, kv := range kvs {
 				kv.StreamId = r.streamIdFor(pk.Attr)
 			}
@@ -665,7 +676,7 @@ func (r *reducer) toList(req *encodeRequest) {
 			}
 		} else {
 			kv := posting.MarshalPostingList(pl, nil)
-			codec.FreePack(pl.Pack)
+			// No need to FreePack here, because we are reusing alloc.
 
 			kv.Key = y.Copy(currentKey)
 			kv.Version = writeVersionTs
@@ -690,7 +701,7 @@ func (r *reducer) toList(req *encodeRequest) {
 
 			if kvBuf.LenNoPadding() > 256<<20 {
 				req.listCh <- kvBuf
-				kvBuf = z.NewBuffer(260 << 20)
+				kvBuf = z.NewBuffer(260<<20, "Reducer.Buffer.KVBuffer")
 			}
 		}
 		end = next

dgraph/cmd/debug/run.go

@@ -472,7 +472,7 @@ func rollupKey(db *badger.DB) {
 	pl, err := posting.ReadPostingList(item.KeyCopy(nil), itr)
 	x.Check(err)
 
-	alloc := z.NewAllocator(32 << 20)
+	alloc := z.NewAllocator(32<<20, "Debug.RollupKey")
 	defer alloc.Release()
 
 	kvs, err := pl.Rollup(alloc)

go.mod

@@ -16,12 +16,12 @@ require (
 	github.com/OneOfOne/xxhash v1.2.5 // indirect
 	github.com/blevesearch/bleve v1.0.13
 	github.com/codahale/hdrhistogram v0.0.0-20161010025455-3a0bb77429bd
-	github.com/dgraph-io/badger/v3 v3.2011.2-0.20210210142907-44c9230e5a66
+	github.com/dgraph-io/badger/v3 v3.2011.2-0.20210323103207-baadc01dca6e
 	github.com/dgraph-io/dgo/v200 v200.0.0-20200805103119-a3544c464dd6
 	github.com/dgraph-io/gqlgen v0.13.2
 	github.com/dgraph-io/gqlparser/v2 v2.1.5
 	github.com/dgraph-io/graphql-transport-ws v0.0.0-20200916064635-48589439591b
-	github.com/dgraph-io/ristretto v0.0.4-0.20210122082011-bb5d392ed82d
+	github.com/dgraph-io/ristretto v0.0.4-0.20210309073149-3836124cdc5a
 	github.com/dgrijalva/jwt-go v3.2.0+incompatible
 	github.com/dgrijalva/jwt-go/v4 v4.0.0-preview1
 	github.com/dgryski/go-farm v0.0.0-20200201041132-a6ae2369ad13
@@ -51,6 +51,7 @@ require (
 	github.com/prometheus/client_golang v0.9.3
 	github.com/prometheus/common v0.4.1
 	github.com/prometheus/procfs v0.0.0-20190517135640-51af30a78b0e // indirect
+	github.com/sirupsen/logrus v1.6.0 // indirect
 	github.com/soheilhy/cmux v0.1.4
 	github.com/spf13/cast v1.3.0
 	github.com/spf13/cobra v0.0.5