Don't reset transactionID when reopening TCP connection

When the client got a IO timeout the timer were never reset
Adjusting timeouts

client.go

@@ -5,29 +5,65 @@ import (
 	"fmt"
 	"io"
 	"net"
+	"sync"
 	"time"
 )
 
 type Mbclient struct {
 	transactionCounter uint16
-	conn               net.Conn
+	address            string
 	header             [7]byte
 	unit               uint8
+	conn               net.Conn
+	t                  *time.Timer
+	keepAliveDuration  time.Duration
+	timeOut            time.Duration
+	wg                 sync.WaitGroup
 }
 
-func New(address string, unit uint8) (*Mbclient, error) {
+func New(Address string, Unit uint8, KeepAlive, TimeOut time.Duration) (*Mbclient, error) {
-	var err error
 
 	c := new(Mbclient)
-	c.conn, err = net.Dial("tcp", address)
+	c.address = Address
-	if err != nil {
+	c.unit = Unit
-		return nil, err
+	c.t = time.NewTimer(0)
-	}
+	<-c.t.C
-	c.unit = unit
+	c.keepAliveDuration = KeepAlive
+	c.timeOut = TimeOut
 	return c, nil
 }
 
+func (m *Mbclient) closer() {
+
+	<-m.t.C
+	m.conn.Close()
+	//m.transactionCounter = 0
+	m.wg.Done()
+}
+
+// closeConn can be called when a tcp communication failure require the current
+// connection to be closed, it should only be used when closer routine is running
+// and the timer is stopped.
+func (m *Mbclient) closeConn() {
+	m.t.Reset(0)
+	m.wg.Wait()
+}
+
 func (m *Mbclient) ReadRegisters(first uint16, numRegs uint16) ([]uint16, error) {
+	var err error
+	// If The timer is expired, conn is closed and needs to be reopened
+	if !m.t.Stop() {
+		// Wait for closer to exit to mitigate race condiion
+		// between closer routine and this code path
+		m.wg.Wait()
+		m.conn, err = net.DialTimeout("tcp", m.address, m.timeOut)
+		if err != nil {
+			return nil, err
+		}
+		m.wg.Add(1)
+		go m.closer()
+	}
+
 	const requestLength = 12
 	m.transactionCounter++
 	req := make([]byte, requestLength)
@@ -41,30 +77,53 @@ func (m *Mbclient) ReadRegisters(first uint16, numRegs uint16) ([]uint16, error)
 	req[7] = 3 //FunctionCode
 	binary.BigEndian.PutUint16(req[8:10], first-1)
 	binary.BigEndian.PutUint16(req[10:12], numRegs)
-	m.conn.SetDeadline(time.Now().Add(10 * time.Second))
+	m.conn.SetDeadline(time.Now().Add(5 * time.Second))
+
 	byteswritten, err := m.conn.Write(req)
 	if err != nil {
+		m.closeConn()
 		return nil, err
 	}
 	if byteswritten != requestLength {
-		return nil, fmt.Errorf("Failed to send request")
+		m.closeConn()
+		return nil, fmt.Errorf("failed to send request")
 	}
-	m.conn.SetDeadline(time.Now().Add(10 * time.Second))
+	m.conn.SetDeadline(time.Now().Add(m.timeOut))
 	_, err = io.ReadFull(m.conn, m.header[:])
+	if err != nil {
+		m.closeConn()
+		return nil, err
+	}
 	responseHeader.unMarshal(m.header)
 	expectedDataLength := responseHeader.length - 1
 
+	if m.transactionCounter != responseHeader.transactionID {
+		m.closeConn()
+		return nil, fmt.Errorf("modbus transaction mismatch %v != %v", m.transactionCounter, responseHeader.transactionID)
+	}
 	response := make([]byte, expectedDataLength)
-	_, err = m.conn.Read(response)
+	m.conn.SetDeadline(time.Now().Add(m.timeOut))
+	bytesRead, err := io.ReadFull(m.conn, response)
+	if err != nil {
+		m.closeConn()
+		return nil, err
+	}
+	if uint16(bytesRead) != expectedDataLength {
+		m.t.Reset(m.keepAliveDuration)
+		return nil, fmt.Errorf("failed to read complete package")
+	}
 
 	err = mbpayload.unMarshal(response)
 	if mbpayload.functionCode != 3 {
-		return nil, fmt.Errorf("modbus exception %v", mbpayload.functionCode&0x7F)
+		m.t.Reset(m.keepAliveDuration)
+		return nil, fmt.Errorf("modbus exception %v req: %x", mbpayload.functionCode&0x7F, req)
 	}
 	if err != nil {
+		m.t.Reset(m.keepAliveDuration)
 		return nil, err
 	}
 
+	m.t.Reset(m.keepAliveDuration)
 	return mbpayload.registers, nil
 }
 
@@ -75,10 +134,6 @@ type mbapHeader struct {
 	unitIdentifier     byte
 }
 
-// func (m *mbapHeader) marshalBinary() ([]byte, error) {
-
-// }
-
 func (m *mbapHeader) unMarshal(data [7]byte) error {
 	m.transactionID = binary.BigEndian.Uint16(data[0:2])
 	m.protocolIdentifier = binary.BigEndian.Uint16(data[2:4])
@@ -98,7 +153,7 @@ func (d *mbPDU) unMarshal(data []byte) error {
 	d.functionCode = data[0]
 	d.length = data[1]
 	if d.length+2 != uint8(len(data)) {
-		return fmt.Errorf("Lenght mismatch in modbus payload")
+		return fmt.Errorf("length mismatch in modbus payload")
 	}
 	d.registers = make([]uint16, d.length/2)
 	var n uint8

client_test.go

@@ -2,12 +2,14 @@ package modbustcpclient
 
 import (
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 )
 
-func TestReadOneRegister(t *testing.T) {
+func TestReadOneRegisterKeepAlive(t *testing.T) {
-	c, err := New("192.168.0.154:502", 1)
+	c, err := New("IAM_248000012514.solver.nu:502", 1, 100*time.Millisecond, 5*time.Second)
+	t.Log("Connect")
 	assert.NoError(t, err)
 	for n := 0; n < 5; n++ {
 		res, err := c.ReadRegisters(12401, 2)
@@ -30,5 +32,64 @@ func TestReadOneRegister(t *testing.T) {
 		assert.Len(t, res, 1)
 		t.Log(res)
 	}
-
+	time.Sleep(1 * time.Second)
+}
+
+func TestReadOneRegisterShortKeepAlive(t *testing.T) {
+	c, err := New("IAM_248000012514.solver.nu:502", 1, 10*time.Nanosecond, 5*time.Second)
+	t.Log("Connect")
+	assert.NoError(t, err)
+	for n := 0; n < 5; n++ {
+		res, err := c.ReadRegisters(12401, 2)
+		assert.NoError(t, err)
+		assert.Len(t, res, 2)
+		t.Log(res)
+		time.Sleep(100 * time.Millisecond)
+
+		res, err = c.ReadRegisters(12102, 2)
+		assert.NoError(t, err)
+		assert.Len(t, res, 2)
+		t.Log(res)
+		time.Sleep(100 * time.Millisecond)
+
+		res, err = c.ReadRegisters(12544, 1)
+		assert.NoError(t, err)
+		assert.Len(t, res, 1)
+		t.Log(float32(res[0]) / 10)
+		time.Sleep(100 * time.Millisecond)
+
+		res, err = c.ReadRegisters(12136, 1)
+		assert.NoError(t, err)
+		assert.Len(t, res, 1)
+		t.Log(res)
+		time.Sleep(100 * time.Millisecond)
+	}
+	time.Sleep(1 * time.Second)
+}
+
+func TestReadALot(t *testing.T) {
+	c, err := New("IAM_248000012514.solver.nu:502", 1, 100*time.Millisecond, 5*time.Second)
+	t.Log("Connect")
+	assert.NoError(t, err)
+	for n := 0; n < 500; n++ {
+		t.Log(n)
+		_, err := c.ReadRegisters(12401, 2)
+		if err != nil {
+			t.Log(err)
+		}
+		_, err = c.ReadRegisters(12102, 2)
+		if err != nil {
+			t.Log(err)
+		}
+
+		_, err = c.ReadRegisters(12544, 1)
+		if err != nil {
+			t.Log(err)
+		}
+		_, err = c.ReadRegisters(12136, 1)
+		if err != nil {
+			t.Log(err)
+		}
+	}
+	time.Sleep(200 * time.Millisecond)
 }