succd: implement auto-restarting of MODBUS connection in case of network loss

succbone/succd/modbus.go

@@ -2,6 +2,7 @@ package main
 
 import (
 	"context"
+	"net"
 	"time"
 
 	"github.com/simonvetter/modbus"
@@ -35,10 +36,17 @@ func (d *daemon) modbusConnect() error {
 	return nil
 }
 
+func (d *daemon) modbusRestart() error {
+	d.modbusClient.Close()
+	return d.modbusClient.Open()
+}
+
 // There are currently two devices connected to the modbus.
 // The first one (slave 1) is a temperature/humidity sensor.
 // The second one (slave 2) is a PTA8D08 transmitter
-func (d *daemon) modbusUpdate() {
+//
+// Returns whether modbus should restart (only in case of an underlying network error)
+func (d *daemon) modbusUpdate() bool {
 	var err error
 
 	// Switch to slave 1 (BTA1)
@@ -48,6 +56,9 @@ func (d *daemon) modbusUpdate() {
 	var registersBTA1 []uint16 // temperature, humidity
 	registersBTA1, err = d.modbusClient.ReadRegisters(1, 2, modbus.INPUT_REGISTER)
 	if err != nil {
+		if _, ok := err.(net.Error); ok {
+			return true
+		}
 		klog.Warningf("error while reading registers from BTA1 %v", err)
 	} else if len(registersBTA1) != 2 {
 		klog.Warningf("expected two registers from modbus slave 1, but got %d", len(registersBTA1))
@@ -68,6 +79,9 @@ func (d *daemon) modbusUpdate() {
 	var registersKEC2 []uint16 // temperatures dp
 	registersKEC2, err = d.modbusClient.ReadRegisters(0, 3, modbus.HOLDING_REGISTER)
 	if err != nil {
+		if _, ok := err.(net.Error); ok {
+			return true
+		}
 		klog.Warningf("error while reading registers from KEC2 %v", err)
 	} else if len(registersKEC2) != 3 {
 		klog.Warningf("expected three registers from modbus slave 2, but got %d", len(registersKEC2))
@@ -88,6 +102,9 @@ func (d *daemon) modbusUpdate() {
 
 	digitalInputRegisters, err = d.modbusClient.ReadRegisters(0x81, 8, modbus.HOLDING_REGISTER)
 	if err != nil {
+		if _, ok := err.(net.Error); ok {
+			return true
+		}
 		klog.Warningf("error while reading digital inputs from KEC1 %v", err)
 	} else {
 		// Convert MODBUS words into bools
@@ -140,8 +157,13 @@ func (d *daemon) modbusUpdate() {
 
 	err = d.modbusClient.WriteRegisters(0x01, registerValuesKEC1[:])
 	if err != nil {
+		if _, ok := err.(net.Error); ok {
+			return true
+		}
 		klog.Warningf("error while updating registers %v", err)
 	}
+
+	return false
 }
 
 // Call modbusUpdate every 100 milliseconds
@@ -151,7 +173,16 @@ func (d *daemon) modbusProcess(ctx context.Context) {
 		case <-ctx.Done():
 			return
 		default:
-			d.modbusUpdate()
+			shouldRestart := d.modbusUpdate()
+			// the modbus library does not reopen the tcp socket in case of
+			// a connection loss.
+			if shouldRestart {
+				klog.Infof("restarting modbus connection...")
+				err := d.modbusRestart()
+				if err != nil {
+					klog.Warningf("failed to restart modbus %v", err)
+				}
+			}
 			time.Sleep(time.Millisecond * 100)
 		}
 	}