[stanza] Use buffer pool for the read buffers to limit allocations

.chloggen/pool-bufs-factory.yaml

@@ -0,0 +1,27 @@
+# Use this changelog template to create an entry for release notes.
+
+# One of 'breaking', 'deprecation', 'new_component', 'enhancement', 'bug_fix'
+change_type: enhancement
+
+# The name of the component, or a single word describing the area of concern, (e.g. filelogreceiver)
+component: pkg/stanza
+
+# A brief description of the change.  Surround your text with quotes ("") if it needs to start with a backtick (`).
+note: Use buffer pool for the read buffers to limit allocations"
+
+# Mandatory: One or more tracking issues related to the change. You can use the PR number here if no issue exists.
+issues: [39373]
+
+# (Optional) One or more lines of additional information to render under the primary note.
+# These lines will be padded with 2 spaces and then inserted directly into the document.
+# Use pipe (|) for multiline entries.
+subtext:
+
+# If your change doesn't affect end users or the exported elements of any package,
+# you should instead start your pull request title with [chore] or use the "Skip Changelog" label.
+# Optional: The change log or logs in which this entry should be included.
+# e.g. '[user]' or '[user, api]'
+# Include 'user' if the change is relevant to end users.
+# Include 'api' if there is a change to a library API.
+# Default: '[user]'
+change_logs: []

pkg/stanza/fileconsumer/benchmark_test.go

@@ -9,6 +9,7 @@ import (
 	"os"
 	"path/filepath"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -161,6 +162,7 @@ func BenchmarkFileInput(b *testing.B) {
 
 	for _, bench := range cases {
 		b.Run(bench.name, func(b *testing.B) {
+			b.ReportAllocs()
 			rootDir := b.TempDir()
 
 			var files []*os.File
@@ -228,3 +230,110 @@ func BenchmarkFileInput(b *testing.B) {
 		})
 	}
 }
+
+func BenchmarkConsumeFiles(b *testing.B) {
+	cases := []fileInputBenchmark{
+		{
+			name: "Single",
+			paths: []string{
+				"file0.log",
+			},
+			config: func() *Config {
+				cfg := NewConfig()
+				cfg.Include = []string{
+					"file0.log",
+				}
+				cfg.MaxLogSize = 1 * 1024 * 1024
+				cfg.InitialBufferSize = 1 * 1024 * 1024
+				cfg.FingerprintSize = fingerprint.DefaultSize / 10
+				return cfg
+			},
+		},
+		{
+			name: "Multiple",
+			paths: func() []string {
+				paths := make([]string, 100)
+				for i := range paths {
+					paths[i] = fmt.Sprintf("file%d.log", i)
+				}
+				return paths
+			}(),
+			config: func() *Config {
+				cfg := NewConfig()
+				cfg.Include = []string{"file*.log"}
+				cfg.Encoding = ""
+				cfg.FingerprintSize = fingerprint.DefaultSize / 10
+				cfg.MaxLogSize = 1 * 1024 * 1024
+				cfg.InitialBufferSize = 1 * 1024 * 1024
+				cfg.MaxConcurrentFiles = 10
+				return cfg
+			},
+		},
+	}
+
+	// Pregenerate some lines which we can write to the files
+	// to avoid measuring the time it takes to generate them
+	// and to reduce the amount of syscalls in the benchmark.
+	uniqueLines := 10
+	severalLines := ""
+	for i := 0; i < uniqueLines; i++ {
+		severalLines += string(filetest.TokenWithLength(999)) + "\n"
+	}
+
+	for _, bench := range cases {
+		b.Run(bench.name, func(b *testing.B) {
+			rootDir := b.TempDir()
+
+			var consumePaths []string
+			var files []*os.File
+			for _, path := range bench.paths {
+				consumePath := filepath.Join(rootDir, path)
+				consumePaths = append(consumePaths, consumePath)
+				f := filetest.OpenFile(b, consumePath)
+				files = append(files, f)
+				// Initialize the file to ensure a unique fingerprint
+				_, err := f.WriteString(f.Name() + "\n")
+				require.NoError(b, err)
+				for i := 0; i < b.N; i++ {
+					_, err := f.WriteString(severalLines)
+					require.NoError(b, err)
+				}
+				require.NoError(b, f.Sync())
+			}
+
+			cfg := bench.config()
+			for i, inc := range cfg.Include {
+				cfg.Include[i] = filepath.Join(rootDir, inc)
+			}
+			cfg.StartAt = "beginning"
+			// Use a long poll so that we don't trigger it.
+			cfg.PollInterval = 1 * time.Hour
+
+			doneChan := make(chan bool, len(files))
+			numTokens := &atomic.Int64{}
+			callback := func(_ context.Context, tokens [][]byte, _ map[string]any, _ int64) error {
+				if numTokens.Add(int64(len(tokens))) == int64(len(files)*(b.N*uniqueLines+1)) {
+					close(doneChan)
+				}
+				return nil
+			}
+			set := componenttest.NewNopTelemetrySettings()
+			op, err := cfg.Build(set, callback)
+			require.NoError(b, err)
+
+			require.NoError(b, op.Start(testutil.NewUnscopedMockPersister()))
+			defer func() {
+				require.NoError(b, op.Stop())
+			}()
+
+			b.ReportAllocs()
+			b.ResetTimer()
+			for len(consumePaths) > op.maxBatchFiles {
+				op.consume(context.Background(), consumePaths[:op.maxBatchFiles])
+				consumePaths = consumePaths[op.maxBatchFiles:]
+			}
+			op.consume(context.Background(), consumePaths)
+			<-doneChan
+		})
+	}
+}

pkg/stanza/fileconsumer/config.go

@@ -152,7 +152,7 @@ func (c Config) Build(set component.TelemetrySettings, emit emit.Callback, opts
 	}
 
 	set.Logger = set.Logger.With(zap.String("component", "fileconsumer"))
-	readerFactory := reader.Factory{
+	readerFactory := &reader.Factory{
 		TelemetrySettings:       set,
 		FromBeginning:           startAtBeginning,
 		FingerprintSize:         int(c.FingerprintSize),

pkg/stanza/fileconsumer/file.go

@@ -27,7 +27,7 @@ type Manager struct {
 	wg     sync.WaitGroup
 	cancel context.CancelFunc
 
-	readerFactory reader.Factory
+	readerFactory *reader.Factory
 	fileMatcher   *matcher.Matcher
 	tracker       tracker.Tracker
 	noTracking    bool

pkg/stanza/fileconsumer/internal/reader/factory.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"sync"
 	"time"
 
 	"go.opentelemetry.io/collector/component"
@@ -34,6 +35,7 @@ type Factory struct {
 	HeaderConfig            *header.Config
 	FromBeginning           bool
 	FingerprintSize         int
+	BufPool                 sync.Pool
 	InitialBufferSize       int
 	MaxLogSize              int
 	Encoding                encoding.Encoding
@@ -75,6 +77,7 @@ func (f *Factory) NewReaderFromMetadata(file *os.File, m *Metadata) (r *Reader,
 		file:              file,
 		fileName:          file.Name(),
 		fingerprintSize:   f.FingerprintSize,
+		bufPool:           &f.BufPool,
 		initialBufferSize: f.InitialBufferSize,
 		maxLogSize:        f.MaxLogSize,
 		decoder:           f.Encoding.NewDecoder(),

pkg/stanza/fileconsumer/internal/reader/reader.go

@@ -8,8 +8,10 @@ import (
 	"compress/gzip"
 	"context"
 	"errors"
+	"fmt"
 	"io"
 	"os"
+	"sync"
 
 	"go.opentelemetry.io/collector/component"
 	"go.uber.org/zap"
@@ -42,6 +44,7 @@ type Reader struct {
 	file                   *os.File
 	reader                 io.Reader
 	fingerprintSize        int
+	bufPool                *sync.Pool
 	initialBufferSize      int
 	maxLogSize             int
 	headerSplitFunc        bufio.SplitFunc
@@ -116,7 +119,9 @@ func (r *Reader) ReadToEnd(ctx context.Context) {
 }
 
 func (r *Reader) readHeader(ctx context.Context) (doneReadingFile bool) {
-	s := scanner.New(r, r.maxLogSize, r.initialBufferSize, r.Offset, r.headerSplitFunc)
+	bufPtr := r.getBufPtrFromPool()
+	defer r.bufPool.Put(bufPtr)
+	s := scanner.New(r, r.maxLogSize, *bufPtr, r.Offset, r.headerSplitFunc)
 
 	// Read the tokens from the file until no more header tokens are found or the end of file is reached.
 	for {
@@ -176,15 +181,19 @@ func (r *Reader) readHeader(ctx context.Context) (doneReadingFile bool) {
 }
 
 func (r *Reader) readContents(ctx context.Context) {
-	// Create the scanner to read the contents of the file.
-	bufferSize := r.initialBufferSize
-	if r.TokenLenState.MinimumLength > bufferSize {
+	var buf []byte
+	fmt.Println(r.fileName)
+	if r.TokenLenState.MinimumLength <= r.initialBufferSize {
+		bufPtr := r.getBufPtrFromPool()
+		buf = *bufPtr
+		defer r.bufPool.Put(bufPtr)
+	} else {
 		// If we previously saw a potential token larger than the default buffer,
-		// size the buffer to be at least one byte larger so we can see if there's more data
-		bufferSize = r.TokenLenState.MinimumLength + 1
+		// size the buffer to be at least one byte larger so we can see if there's more data.
+		// Usually, expect this to be a rare event so that we don't bother pooling this special buffer size.
+		buf = make([]byte, 0, r.TokenLenState.MinimumLength+1)
 	}
-
-	s := scanner.New(r, r.maxLogSize, bufferSize, r.Offset, r.contentSplitFunc)
+	s := scanner.New(r, r.maxLogSize, buf, r.Offset, r.contentSplitFunc)
 
 	tokenBodies := make([][]byte, r.maxBatchSize)
 	numTokensBatched := 0
@@ -319,3 +328,12 @@ func (r *Reader) updateFingerprint() {
 	}
 	r.Fingerprint = refreshedFingerprint
 }
+
+func (r *Reader) getBufPtrFromPool() *[]byte {
+	bufP := r.bufPool.Get()
+	if bufP == nil {
+		buf := make([]byte, 0, r.initialBufferSize)
+		return &buf
+	}
+	return bufP.(*[]byte)
+}

pkg/stanza/fileconsumer/internal/scanner/scanner.go

@@ -20,9 +20,9 @@ type Scanner struct {
 }
 
 // New creates a new positional scanner
-func New(r io.Reader, maxLogSize int, bufferSize int, startOffset int64, splitFunc bufio.SplitFunc) *Scanner {
+func New(r io.Reader, maxLogSize int, buf []byte, startOffset int64, splitFunc bufio.SplitFunc) *Scanner {
 	s := &Scanner{Scanner: bufio.NewScanner(r), pos: startOffset}
-	s.Buffer(make([]byte, 0, bufferSize), maxLogSize)
+	s.Buffer(buf, maxLogSize)
 	scanFunc := func(data []byte, atEOF bool) (advance int, token []byte, err error) {
 		advance, token, err = splitFunc(data, atEOF)
 		s.pos += int64(advance)

pkg/stanza/fileconsumer/internal/scanner/scanner_test.go

@@ -77,7 +77,7 @@ func TestScanner(t *testing.T) {
 
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
-			scanner := New(bytes.NewReader(tc.stream), tc.maxSize, tc.bufferSize, tc.startOffset, simpleSplit(tc.delimiter))
+			scanner := New(bytes.NewReader(tc.stream), tc.maxSize, make([]byte, 0, tc.bufferSize), tc.startOffset, simpleSplit(tc.delimiter))
 			for i, p := 0, 0; scanner.Scan(); i++ {
 				assert.NoError(t, scanner.Error())
 
@@ -117,12 +117,12 @@ func (r *errReader) Read([]byte) (n int, err error) {
 
 func TestScannerError(t *testing.T) {
 	reader := &errReader{err: bufio.ErrTooLong}
-	scanner := New(reader, 100, 100, 0, simpleSplit([]byte("\n")))
+	scanner := New(reader, 100, make([]byte, 0, 100), 0, simpleSplit([]byte("\n")))
 	assert.False(t, scanner.Scan())
 	assert.EqualError(t, scanner.Error(), "log entry too large")
 
 	reader = &errReader{err: errors.New("some err")}
-	scanner = New(reader, 100, 100, 0, simpleSplit([]byte("\n")))
+	scanner = New(reader, 100, make([]byte, 0, 100), 0, simpleSplit([]byte("\n")))
 	assert.False(t, scanner.Scan())
 	assert.EqualError(t, scanner.Error(), "scanner error: some err")
 }