Preallocate p.stacks() before calling runtime.GoroutineProfile()

fgprof.go

@@ -20,7 +20,7 @@ func Start(w io.Writer, format Format) func() error {
 	ticker := time.NewTicker(time.Second / hz)
 	stopCh := make(chan struct{})
 
-	prof := &profiler{}
+	prof := newProfiler()
 	stackCounts := stackCounter{}
 
 	go func() {
@@ -50,6 +50,16 @@ type profiler struct {
 	selfFrame *runtime.Frame
 }
 
+// newProfiler returns a new profiler with a pre-allocated stacks field. This
+// is slightly faster than discovering its size during the first
+// GoroutineProfile() call.
+func newProfiler() *profiler {
+	n := runtime.NumGoroutine()
+	return &profiler{
+		stacks: make([]runtime.StackRecord, int(float64(n)*1.1)),
+	}
+}
+
 // GoroutineProfile returns the stacks of all goroutines currently managed by
 // the scheduler. This includes both goroutines that are currently running
 // (On-CPU), as well as waiting (Off-CPU).

fgprof_test.go

@@ -3,6 +3,7 @@ package fgprof
 import (
 	"bytes"
 	"fmt"
+	"runtime"
 	"strings"
 	"testing"
 	"time"
@@ -26,7 +27,7 @@ func TestStart(t *testing.T) {
 }
 
 func BenchmarkProfiler(b *testing.B) {
-	prof := &profiler{}
+	prof := newProfiler()
 	for i := 0; i < b.N; i++ {
 		prof.GoroutineProfile()
 	}
@@ -38,8 +39,7 @@ func BenchmarkProfilerGoroutines(b *testing.B) {
 		name := fmt.Sprintf("%d goroutines", g)
 
 		b.Run(name, func(b *testing.B) {
-			prof := &profiler{}
-			initalRoutines := len(prof.GoroutineProfile())
+			initalRoutines := runtime.NumGoroutine()
 
 			readyCh := make(chan struct{})
 			stopCh := make(chan struct{})
@@ -51,6 +51,9 @@ func BenchmarkProfilerGoroutines(b *testing.B) {
 				<-readyCh
 			}
 
+			// allocate profiler after some goroutines has been generated
+			prof := newProfiler()
+
 			b.ResetTimer()
 			for i := 0; i < b.N; i++ {
 				stacks := prof.GoroutineProfile()
@@ -78,7 +81,7 @@ func BenchmarkProfilerGoroutines(b *testing.B) {
 }
 
 func BenchmarkStackCounter(b *testing.B) {
-	prof := &profiler{}
+	prof := newProfiler()
 	stacks := prof.GoroutineProfile()
 	sc := stackCounter{}
 	b.ResetTimer()