More stable e2e test harness, better for benchmarking (#1702)

overlay/tun_tester.go

@@ -15,6 +15,7 @@ import (
 	"github.com/gaissmai/bart"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/routing"
+	"github.com/slackhq/nebula/udp"
 )
 
 type TestTun struct {
@@ -54,9 +55,12 @@ func newTunFromFd(_ *config.C, _ *slog.Logger, _ int, _ []netip.Prefix) (*TestTu
 	return nil, fmt.Errorf("newTunFromFd not supported")
 }
 
-// Send will place a byte array onto the receive queue for nebula to consume
+// Send will place a byte array onto the receive queue for nebula to consume.
 // These are unencrypted ip layer frames destined for another nebula node.
-// packets should exit the udp side, capture them with udpConn.Get
+// packets should exit the udp side, capture them with udpConn.Get.
+//
+// Send copies the input via the freelist, so the caller is free to mutate
+// or reuse it after the call returns.
 func (t *TestTun) Send(packet []byte) {
 	if t.closed.Load() {
 		return
@@ -65,7 +69,9 @@ func (t *TestTun) Send(packet []byte) {
 	if t.l.Enabled(context.Background(), slog.LevelDebug) {
 		t.l.Debug("Tun receiving injected packet", "dataLen", len(packet))
 	}
-	t.rxPackets <- packet
+	buf := acquireTunBuf(len(packet))
+	copy(buf, packet)
+	t.rxPackets <- buf
 }
 
 // Get will pull an unencrypted ip layer frame from the transmit queue
@@ -110,12 +116,44 @@ func (t *TestTun) Write(b []byte) (n int, err error) {
 		return 0, io.ErrClosedPipe
 	}
 
-	packet := make([]byte, len(b), len(b))
+	packet := acquireTunBuf(len(b))
 	copy(packet, b)
 	t.TxPackets <- packet
 	return len(b), nil
 }
 
+// ReleaseTunBuf returns a slice from TxPackets to the harness freelist, don't use the bytes after the call.
+// Channel-backed instead of sync.Pool because putting a []byte in a sync.Pool escapes the slice header to heap.
+func ReleaseTunBuf(b []byte) {
+	if b == nil {
+		return
+	}
+	select {
+	case tunBufFreelist <- b:
+	default:
+		// Freelist full; drop the buffer for the GC.
+	}
+}
+
+// tunBufFreelist retains the backing arrays for TestTun.Write so steady-state allocation drops to zero once the
+// freelist has saturated for the current MTU.
+var tunBufFreelist = make(chan []byte, 64)
+
+func acquireTunBuf(n int) []byte {
+	var b []byte
+	select {
+	case b = <-tunBufFreelist:
+	default:
+		b = make([]byte, 0, udp.MTU)
+	}
+	if cap(b) < n {
+		b = make([]byte, n)
+	} else {
+		b = b[:n]
+	}
+	return b
+}
+
 func (t *TestTun) Close() error {
 	if t.closed.CompareAndSwap(false, true) {
 		close(t.rxPackets)
@@ -129,8 +167,14 @@ func (t *TestTun) Read(b []byte) (int, error) {
 	if !ok {
 		return 0, os.ErrClosed
 	}
+	n := len(p)
 	copy(b, p)
-	return len(p), nil
+	// Send always pushes a freelist-acquired slice, return it once we've copied the bytes into the caller's buffer.
+	select {
+	case tunBufFreelist <- p:
+	default:
+	}
+	return n, nil
 }
 
 func (t *TestTun) SupportsMultiqueue() bool {