Making more robust in case ping fails.

interface.go

@@ -111,12 +111,29 @@ func (b *AbstractBridge) removeIPandBridgeInt() {
 
 }
 
-func isActive(bridgeip string) bool {
+func (b *AbstractBridge) IsActive() bool {
+
+	defer func() {
+		if r := recover(); r != nil {
+			fmt.Println("An error happened in <WriteNumberToMulticast()>, but Zoreide recovered. ")
+			fmt.Println("Error was: ", r)
+		}
+	}()
+
+	var bridgeip string
+
+	brIp, _, err := net.ParseCIDR(b.BridgeIpCIDR)
+	if err != nil {
+		log.Println("IsActive : problem parsing the IP/CIDR: ", err.Error())
+	} else {
+		bridgeip = brIp.String()
+	}
 
 	log.Println("Check for active IP: ", bridgeip)
 
 	pinger, err := ping.NewPinger(bridgeip)
 	if err != nil {
+		log.Println("Unable to ping address: ", bridgeip)
 		log.Println("Ping error: " + err.Error())
 	}
 	// just in case it doesn't stops alone
@@ -133,3 +150,65 @@ func isActive(bridgeip string) bool {
 	return stats.PacketsRecv == pinger.Count
 
 }
+
+func (b *AbstractBridge) IsAssigned() bool {
+
+	// we want the program to recover in case of issues
+	defer func() {
+		if r := recover(); r != nil {
+
+			fmt.Println("An error happened in <IsAssigned()>, but Zoreide recovered. ")
+			fmt.Println("Error was: ", r)
+		}
+	}()
+
+	var (
+		ief      *net.Interface
+		addrs    []net.Addr
+		ipv4Addr net.IP
+		err      error
+		bridgeip string
+	)
+
+	brIp, _, err := net.ParseCIDR(b.BridgeIpCIDR)
+	if err != nil {
+		log.Println("IsActive : problem parsing the IP/CIDR: ", err.Error())
+		return false
+	} else {
+		bridgeip = brIp.String()
+	}
+
+	interfaceName := b.ExistingInterface
+
+	if ief, err = net.InterfaceByName(interfaceName); err != nil { // get interface
+		log.Printf("Interface %s does not exist or not manageable\n", interfaceName)
+		log.Printf("Error is: %s\n", err.Error())
+		return false
+	}
+	if addrs, err = ief.Addrs(); err != nil { // get addresses
+		log.Printf("Cannot read IPs of interface %s\n", interfaceName)
+		log.Printf("Error is: %s\n", err.Error())
+		return false
+	}
+	for _, addr := range addrs { // get ipv4 address
+		if ipv4Addr = addr.(*net.IPNet).IP.To4(); ipv4Addr != nil {
+			log.Printf("Error reading IPs for interface %s\n", interfaceName)
+			log.Printf("Error is: %s\n", err.Error())
+			break
+		}
+	}
+	if ipv4Addr == nil {
+		log.Printf("interface %s don't have an ipv4 address\n", interfaceName)
+		return false
+	}
+
+	for _, addr := range addrs {
+		if addr.String() == bridgeip {
+			log.Printf("Ip %s is assigned to interface %s", bridgeip, interfaceName)
+			return true
+		}
+	}
+
+	log.Printf("Ip %s is NOT assigned to interface %s", bridgeip, interfaceName)
+	return false
+}

orchestrator.go

@@ -3,7 +3,6 @@ package main
 import (
 	"fmt"
 	"log"
-	"net"
 	"regexp"
 	"slices"
 	"strconv"
@@ -55,6 +54,13 @@ func (mip *AbstractMulticast) WriteNumberToMulticast(br AbstractBridge) {
 
 func (mip *AbstractMulticast) ReadNumberFromMulticast() {
 
+	defer func() {
+		if r := recover(); r != nil {
+			fmt.Println("An error happened in <ReadNumberToMulticast()>, but Zoreide recovered. ")
+			fmt.Println("Error was: ", r)
+		}
+	}()
+
 	log.Println("Initiating reader")
 	buffer := make([]byte, mip.MaxDatagramSize)
 
@@ -74,6 +80,13 @@ func (mip *AbstractMulticast) ReadNumberFromMulticast() {
 
 func (b *AbstractBridge) HierarchyReLocator(entity AbstractMulticast) {
 
+	defer func() {
+		if r := recover(); r != nil {
+			fmt.Println("An error happened in <HierarchyLocator()>, but Zoreide recovered. ")
+			fmt.Println("Error was: ", r)
+		}
+	}()
+
 	log.Println("Inizializing HierarchyManager")
 
 	entity.AddUniqueAndSort(b.hIerarchyNumber)
@@ -101,26 +114,26 @@ func (b *AbstractBridge) HierarchyReLocator(entity AbstractMulticast) {
 			entity.AddUniqueAndSort(brdNumber)
 		}
 
-		brIp, _, err := net.ParseCIDR(b.BridgeIpCIDR)
-		if err != nil {
-			log.Println(err.Error())
-		}
-
 		// finished feeding the new number
 		// if Alpha:
 		if entity.IsAlpha(b.hIerarchyNumber) {
-			if isActive(brIp.String()) {
+			if b.IsAssigned() {
 				log.Println("Still ALPHA. This is ok.")
 			} else {
 				log.Println("I'm the new ALPHA! Get out my path, losers!")
 				b.configureIpAndBridgeUp()
-				log.Println("Ip is active: ", isActive(brIp.String()))
+				log.Println("Ip is active: ", b.IsActive())
 			}
+			// here we manage the case when we are not alpha
 		} else {
 			log.Println("GULP! There is a bigger one, better descalating")
-			if isActive(brIp.String()) {
+			if b.IsAssigned() {
+				log.Println("Start removing the IP from the interface:")
 				b.removeIPandBridgeInt()
-
+				log.Println("Removed.Ip is reachable: ", b.IsActive())
+			} else {
+				log.Println("Nothing to do, since IP is not assigned to the interface: ", b.ExistingInterface)
+				log.Println("Ip is reachable: ", b.IsActive())
 			}
 		}
 
@@ -130,22 +143,40 @@ func (b *AbstractBridge) HierarchyReLocator(entity AbstractMulticast) {
 
 func (b *AbstractBridge) WaitAndClean(entity AbstractMulticast) {
 
-	log.Println("Inizializing HAManager")
+	defer func() {
+		if r := recover(); r != nil {
+			fmt.Println("An error happened in <WaitAndClean()>, but Zoreide recovered. ")
+			fmt.Println("Error was: ", r)
+		}
+	}()
 
-	brIp, _, err := net.ParseCIDR(b.BridgeIpCIDR)
+	log.Println("Inizializing HA-Manager")
-	if err != nil {
-		log.Println(err.Error())
-	}
 
 	for {
 
 		pollTime := len(entity.HierarchyArray) + 1
 		time.Sleep(time.Duration(pollTime) * time.Second)
-		// svuotare l'array e rifare le elezioni
+		// Evitiamo di avere l'IP senza essere alpha
-		if !isActive(string(brIp.String())) {
+		if b.IsAssigned() {
+			if entity.IsAlpha(b.hIerarchyNumber) {
+				log.Println("We are alpha and IP is assigned. All ok")
+			} else {
+				log.Println("Cannot have IP assigned without being alpha")
+				log.Println("Inconsistent situation: start removing the IP from the interface:")
+				b.removeIPandBridgeInt()
+				log.Println("Removed.Ip is still reachable: ", b.IsActive())
+			}
+		}
+
+		if b.IsActive() {
+			log.Println("Ip reachable, cluster OK")
+			continue
+		} else {
+			log.Println("Situation non consistent, restart from green field.")
 			entity.HierarchyArray = entity.HierarchyArray[:0]
 			entity.HierarchyArray = slices.Clip(entity.HierarchyArray)
 			entity.AddUniqueAndSort(b.hIerarchyNumber)
+			log.Println("Restarted from green field: ", entity.HierarchyArray)
 		}
 
 	}