rpc_util: Reuse memory buffer for receiving message

benchmark/benchmain/main.go

@@ -109,6 +109,7 @@ var (
 	clientWriteBufferSize = flags.IntSlice("clientWriteBufferSize", []int{-1}, "Configures the client write buffer size in bytes. If negative, use the default - may be a a comma-separated list")
 	serverReadBufferSize  = flags.IntSlice("serverReadBufferSize", []int{-1}, "Configures the server read buffer size in bytes. If negative, use the default - may be a a comma-separated list")
 	serverWriteBufferSize = flags.IntSlice("serverWriteBufferSize", []int{-1}, "Configures the server write buffer size in bytes. If negative, use the default - may be a a comma-separated list")
+	recvBufferPool        = flags.StringWithAllowedValues("recvBufferPool", recvBufferPoolNil, "Configures the shared receive buffer pool. One of: nil, simple, all", allRecvBufferPools)
 
 	logger = grpclog.Component("benchmark")
 )
@@ -133,6 +134,10 @@ const (
 	networkModeLAN   = "LAN"
 	networkModeWAN   = "WAN"
 	networkLongHaul  = "Longhaul"
+	// Shared recv buffer pool
+	recvBufferPoolNil    = "nil"
+	recvBufferPoolSimple = "simple"
+	recvBufferPoolAll    = "all"
 
 	numStatsBuckets = 10
 	warmupCallCount = 10
@@ -144,6 +149,7 @@ var (
 	allCompModes              = []string{compModeOff, compModeGzip, compModeNop, compModeAll}
 	allToggleModes            = []string{toggleModeOff, toggleModeOn, toggleModeBoth}
 	allNetworkModes           = []string{networkModeNone, networkModeLocal, networkModeLAN, networkModeWAN, networkLongHaul}
+	allRecvBufferPools        = []string{recvBufferPoolNil, recvBufferPoolSimple, recvBufferPoolAll}
 	defaultReadLatency        = []time.Duration{0, 40 * time.Millisecond} // if non-positive, no delay.
 	defaultReadKbps           = []int{0, 10240}                           // if non-positive, infinite
 	defaultReadMTU            = []int{0}                                  // if non-positive, infinite
@@ -321,6 +327,15 @@ func makeClient(bf stats.Features) (testgrpc.BenchmarkServiceClient, func()) {
 	if bf.ServerWriteBufferSize >= 0 {
 		sopts = append(sopts, grpc.WriteBufferSize(bf.ServerWriteBufferSize))
 	}
+	switch bf.RecvBufferPool {
+	case recvBufferPoolNil:
+		// Do nothing.
+	case recvBufferPoolSimple:
+		opts = append(opts, grpc.WithRecvBufferPool(grpc.NewSharedBufferPool()))
+		sopts = append(sopts, grpc.RecvBufferPool(grpc.NewSharedBufferPool()))
+	default:
+		logger.Fatalf("Unknown shared recv buffer pool type: %v", bf.RecvBufferPool)
+	}
 
 	sopts = append(sopts, grpc.MaxConcurrentStreams(uint32(bf.MaxConcurrentCalls+1)))
 	opts = append(opts, grpc.WithTransportCredentials(insecure.NewCredentials()))
@@ -528,6 +543,7 @@ type featureOpts struct {
 	clientWriteBufferSize []int
 	serverReadBufferSize  []int
 	serverWriteBufferSize []int
+	recvBufferPools       []string
 }
 
 // makeFeaturesNum returns a slice of ints of size 'maxFeatureIndex' where each
@@ -572,6 +588,8 @@ func makeFeaturesNum(b *benchOpts) []int {
 			featuresNum[i] = len(b.features.serverReadBufferSize)
 		case stats.ServerWriteBufferSize:
 			featuresNum[i] = len(b.features.serverWriteBufferSize)
+		case stats.RecvBufferPool:
+			featuresNum[i] = len(b.features.recvBufferPools)
 		default:
 			log.Fatalf("Unknown feature index %v in generateFeatures. maxFeatureIndex is %v", i, stats.MaxFeatureIndex)
 		}
@@ -638,6 +656,7 @@ func (b *benchOpts) generateFeatures(featuresNum []int) []stats.Features {
 			ClientWriteBufferSize: b.features.clientWriteBufferSize[curPos[stats.ClientWriteBufferSize]],
 			ServerReadBufferSize:  b.features.serverReadBufferSize[curPos[stats.ServerReadBufferSize]],
 			ServerWriteBufferSize: b.features.serverWriteBufferSize[curPos[stats.ServerWriteBufferSize]],
+			RecvBufferPool:        b.features.recvBufferPools[curPos[stats.RecvBufferPool]],
 		}
 		if len(b.features.reqPayloadCurves) == 0 {
 			f.ReqSizeBytes = b.features.reqSizeBytes[curPos[stats.ReqSizeBytesIndex]]
@@ -708,6 +727,7 @@ func processFlags() *benchOpts {
 			clientWriteBufferSize: append([]int(nil), *clientWriteBufferSize...),
 			serverReadBufferSize:  append([]int(nil), *serverReadBufferSize...),
 			serverWriteBufferSize: append([]int(nil), *serverWriteBufferSize...),
+			recvBufferPools:       setRecvBufferPool(*recvBufferPool),
 		},
 	}
 
@@ -783,6 +803,19 @@ func setCompressorMode(val string) []string {
 	}
 }
 
+func setRecvBufferPool(val string) []string {
+	switch val {
+	case recvBufferPoolNil, recvBufferPoolSimple:
+		return []string{val}
+	case recvBufferPoolAll:
+		return []string{recvBufferPoolNil, recvBufferPoolSimple}
+	default:
+		// This should never happen because a wrong value passed to this flag would
+		// be caught during flag.Parse().
+		return []string{}
+	}
+}
+
 func main() {
 	opts := processFlags()
 	before(opts)

benchmark/stats/stats.go

@@ -56,6 +56,7 @@ const (
 	ClientWriteBufferSize
 	ServerReadBufferSize
 	ServerWriteBufferSize
+	RecvBufferPool
 
 	// MaxFeatureIndex is a place holder to indicate the total number of feature
 	// indices we have. Any new feature indices should be added above this.
@@ -121,6 +122,8 @@ type Features struct {
 	ServerReadBufferSize int
 	// ServerWriteBufferSize is the size of the server write buffer in bytes. If negative, use the default buffer size.
 	ServerWriteBufferSize int
+	// RecvBufferPool represents the shared recv buffer pool used.
+	RecvBufferPool string
 }
 
 // String returns all the feature values as a string.
@@ -139,12 +142,13 @@ func (f Features) String() string {
 	return fmt.Sprintf("networkMode_%v-bufConn_%v-keepalive_%v-benchTime_%v-"+
 		"trace_%v-latency_%v-kbps_%v-MTU_%v-maxConcurrentCalls_%v-%s-%s-"+
 		"compressor_%v-channelz_%v-preloader_%v-clientReadBufferSize_%v-"+
-		"clientWriteBufferSize_%v-serverReadBufferSize_%v-serverWriteBufferSize_%v-",
+		"clientWriteBufferSize_%v-serverReadBufferSize_%v-serverWriteBufferSize_%v-"+
+		"recvBufferPool_%v-",
 		f.NetworkMode, f.UseBufConn, f.EnableKeepalive, f.BenchTime, f.EnableTrace,
 		f.Latency, f.Kbps, f.MTU, f.MaxConcurrentCalls, reqPayloadString,
 		respPayloadString, f.ModeCompressor, f.EnableChannelz, f.EnablePreloader,
 		f.ClientReadBufferSize, f.ClientWriteBufferSize, f.ServerReadBufferSize,
-		f.ServerWriteBufferSize)
+		f.ServerWriteBufferSize, f.RecvBufferPool)
 }
 
 // SharedFeatures returns the shared features as a pretty printable string.
@@ -216,6 +220,8 @@ func (f Features) partialString(b *bytes.Buffer, wantFeatures []bool, sep, delim
 				b.WriteString(fmt.Sprintf("ServerReadBufferSize%v%v%v", sep, f.ServerReadBufferSize, delim))
 			case ServerWriteBufferSize:
 				b.WriteString(fmt.Sprintf("ServerWriteBufferSize%v%v%v", sep, f.ServerWriteBufferSize, delim))
+			case RecvBufferPool:
+				b.WriteString(fmt.Sprintf("RecvBufferPool%v%v%v", sep, f.RecvBufferPool, delim))
 			default:
 				log.Fatalf("Unknown feature index %v. maxFeatureIndex is %v", i, MaxFeatureIndex)
 			}

dialoptions.go

@@ -78,6 +78,7 @@ type dialOptions struct {
 	defaultServiceConfigRawJSON *string
 	resolvers                   []resolver.Builder
 	idleTimeout                 time.Duration
+	recvBufferPool              SharedBufferPool
 }
 
 // DialOption configures how we set up the connection.
@@ -628,7 +629,8 @@ func defaultDialOptions() dialOptions {
 			ReadBufferSize:  defaultReadBufSize,
 			UseProxy:        true,
 		},
-		idleTimeout: 30 * time.Minute,
+		idleTimeout:    30 * time.Minute,
+		recvBufferPool: nopBufferPool{},
 	}
 }
 
@@ -677,3 +679,24 @@ func WithIdleTimeout(d time.Duration) DialOption {
 		o.idleTimeout = d
 	})
 }
+
+// WithRecvBufferPool returns a DialOption that configures the ClientConn
+// to use the provided shared buffer pool for parsing incoming messages. Depending
+// on the application's workload, this could result in reduced memory allocation.
+//
+// If you are unsure about how to implement a memory pool but want to utilize one,
+// begin with grpc.NewSharedBufferPool.
+//
+// Note: The shared buffer pool feature will not be active if any of the following
+// options are used: WithStatsHandler, EnableTracing, or binary logging. In such
+// cases, the shared buffer pool will be ignored.
+//
+// # Experimental
+//
+// Notice: This API is EXPERIMENTAL and may be changed or removed in a
+// later release.
+func WithRecvBufferPool(bufferPool SharedBufferPool) DialOption {
+	return newFuncDialOption(func(o *dialOptions) {
+		o.recvBufferPool = bufferPool
+	})
+}

rpc_util.go

@@ -577,6 +577,9 @@ type parser struct {
 	// The header of a gRPC message. Find more detail at
 	// https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md
 	header [5]byte
+
+	// recvBufferPool is the pool of shared receive buffers.
+	recvBufferPool SharedBufferPool
 }
 
 // recvMsg reads a complete gRPC message from the stream.
@@ -610,9 +613,7 @@ func (p *parser) recvMsg(maxReceiveMessageSize int) (pf payloadFormat, msg []byt
 	if int(length) > maxReceiveMessageSize {
 		return 0, nil, status.Errorf(codes.ResourceExhausted, "grpc: received message larger than max (%d vs. %d)", length, maxReceiveMessageSize)
 	}
-	// TODO(bradfitz,zhaoq): garbage. reuse buffer after proto decoding instead
-	// of making it for each message:
-	msg = make([]byte, int(length))
+	msg = p.recvBufferPool.Get(int(length))
 	if _, err := p.r.Read(msg); err != nil {
 		if err == io.EOF {
 			err = io.ErrUnexpectedEOF
@@ -726,12 +727,12 @@ type payloadInfo struct {
 }
 
 func recvAndDecompress(p *parser, s *transport.Stream, dc Decompressor, maxReceiveMessageSize int, payInfo *payloadInfo, compressor encoding.Compressor) ([]byte, error) {
-	pf, d, err := p.recvMsg(maxReceiveMessageSize)
+	pf, buf, err := p.recvMsg(maxReceiveMessageSize)
 	if err != nil {
 		return nil, err
 	}
 	if payInfo != nil {
-		payInfo.compressedLength = len(d)
+		payInfo.compressedLength = len(buf)
 	}
 
 	if st := checkRecvPayload(pf, s.RecvCompress(), compressor != nil || dc != nil); st != nil {
@@ -743,10 +744,10 @@ func recvAndDecompress(p *parser, s *transport.Stream, dc Decompressor, maxRecei
 		// To match legacy behavior, if the decompressor is set by WithDecompressor or RPCDecompressor,
 		// use this decompressor as the default.
 		if dc != nil {
-			d, err = dc.Do(bytes.NewReader(d))
-			size = len(d)
+			buf, err = dc.Do(bytes.NewReader(buf))
+			size = len(buf)
 		} else {
-			d, size, err = decompress(compressor, d, maxReceiveMessageSize)
+			buf, size, err = decompress(compressor, buf, maxReceiveMessageSize)
 		}
 		if err != nil {
 			return nil, status.Errorf(codes.Internal, "grpc: failed to decompress the received message: %v", err)
@@ -757,7 +758,7 @@ func recvAndDecompress(p *parser, s *transport.Stream, dc Decompressor, maxRecei
 			return nil, status.Errorf(codes.ResourceExhausted, "grpc: received message after decompression larger than max (%d vs. %d)", size, maxReceiveMessageSize)
 		}
 	}
-	return d, nil
+	return buf, nil
 }
 
 // Using compressor, decompress d, returning data and size.
@@ -792,15 +793,17 @@ func decompress(compressor encoding.Compressor, d []byte, maxReceiveMessageSize
 // dc takes precedence over compressor.
 // TODO(dfawley): wrap the old compressor/decompressor using the new API?
 func recv(p *parser, c baseCodec, s *transport.Stream, dc Decompressor, m interface{}, maxReceiveMessageSize int, payInfo *payloadInfo, compressor encoding.Compressor) error {
-	d, err := recvAndDecompress(p, s, dc, maxReceiveMessageSize, payInfo, compressor)
+	buf, err := recvAndDecompress(p, s, dc, maxReceiveMessageSize, payInfo, compressor)
 	if err != nil {
 		return err
 	}
-	if err := c.Unmarshal(d, m); err != nil {
+	if err := c.Unmarshal(buf, m); err != nil {
 		return status.Errorf(codes.Internal, "grpc: failed to unmarshal the received message: %v", err)
 	}
 	if payInfo != nil {
-		payInfo.uncompressedBytes = d
+		payInfo.uncompressedBytes = buf
+	} else {
+		p.recvBufferPool.Put(&buf)
 	}
 	return nil
 }

rpc_util_test.go

@@ -65,7 +65,7 @@ func (s) TestSimpleParsing(t *testing.T) {
 		{append([]byte{0, 1, 0, 0, 0}, bigMsg...), nil, bigMsg, compressionNone},
 	} {
 		buf := fullReader{bytes.NewReader(test.p)}
-		parser := &parser{r: buf}
+		parser := &parser{r: buf, recvBufferPool: nopBufferPool{}}
 		pt, b, err := parser.recvMsg(math.MaxInt32)
 		if err != test.err || !bytes.Equal(b, test.b) || pt != test.pt {
 			t.Fatalf("parser{%v}.recvMsg(_) = %v, %v, %v\nwant %v, %v, %v", test.p, pt, b, err, test.pt, test.b, test.err)
@@ -77,7 +77,7 @@ func (s) TestMultipleParsing(t *testing.T) {
 	// Set a byte stream consists of 3 messages with their headers.
 	p := []byte{0, 0, 0, 0, 1, 'a', 0, 0, 0, 0, 2, 'b', 'c', 0, 0, 0, 0, 1, 'd'}
 	b := fullReader{bytes.NewReader(p)}
-	parser := &parser{r: b}
+	parser := &parser{r: b, recvBufferPool: nopBufferPool{}}
 
 	wantRecvs := []struct {
 		pt   payloadFormat

server.go

@@ -174,6 +174,7 @@ type serverOptions struct {
 	maxHeaderListSize     *uint32
 	headerTableSize       *uint32
 	numServerWorkers      uint32
+	recvBufferPool        SharedBufferPool
 }
 
 var defaultServerOptions = serverOptions{
@@ -182,6 +183,7 @@ var defaultServerOptions = serverOptions{
 	connectionTimeout:     120 * time.Second,
 	writeBufferSize:       defaultWriteBufSize,
 	readBufferSize:        defaultReadBufSize,
+	recvBufferPool:        nopBufferPool{},
 }
 var globalServerOptions []ServerOption
 
@@ -552,6 +554,27 @@ func NumStreamWorkers(numServerWorkers uint32) ServerOption {
 	})
 }
 
+// RecvBufferPool returns a ServerOption that configures the server
+// to use the provided shared buffer pool for parsing incoming messages. Depending
+// on the application's workload, this could result in reduced memory allocation.
+//
+// If you are unsure about how to implement a memory pool but want to utilize one,
+// begin with grpc.NewSharedBufferPool.
+//
+// Note: The shared buffer pool feature will not be active if any of the following
+// options are used: StatsHandler, EnableTracing, or binary logging. In such
+// cases, the shared buffer pool will be ignored.
+//
+// # Experimental
+//
+// Notice: This API is EXPERIMENTAL and may be changed or removed in a
+// later release.
+func RecvBufferPool(bufferPool SharedBufferPool) ServerOption {
+	return newFuncServerOption(func(o *serverOptions) {
+		o.recvBufferPool = bufferPool
+	})
+}
+
 // serverWorkerResetThreshold defines how often the stack must be reset. Every
 // N requests, by spawning a new goroutine in its place, a worker can reset its
 // stack so that large stacks don't live in memory forever. 2^16 should allow
@@ -1296,7 +1319,7 @@ func (s *Server) processUnaryRPC(t transport.ServerTransport, stream *transport.
 	if len(shs) != 0 || len(binlogs) != 0 {
 		payInfo = &payloadInfo{}
 	}
-	d, err := recvAndDecompress(&parser{r: stream}, stream, dc, s.opts.maxReceiveMessageSize, payInfo, decomp)
+	d, err := recvAndDecompress(&parser{r: stream, recvBufferPool: s.opts.recvBufferPool}, stream, dc, s.opts.maxReceiveMessageSize, payInfo, decomp)
 	if err != nil {
 		if e := t.WriteStatus(stream, status.Convert(err)); e != nil {
 			channelz.Warningf(logger, s.channelzID, "grpc: Server.processUnaryRPC failed to write status: %v", e)
@@ -1506,7 +1529,7 @@ func (s *Server) processStreamingRPC(t transport.ServerTransport, stream *transp
 		ctx:                   ctx,
 		t:                     t,
 		s:                     stream,
-		p:                     &parser{r: stream},
+		p:                     &parser{r: stream, recvBufferPool: s.opts.recvBufferPool},
 		codec:                 s.getCodec(stream.ContentSubtype()),
 		maxReceiveMessageSize: s.opts.maxReceiveMessageSize,
 		maxSendMessageSize:    s.opts.maxSendMessageSize,