🧱 Introduce etcd for high availability (HA)

pkg/internal/directory/command.go

@@ -18,5 +18,5 @@ type Command struct {
 	// The implementation of the command, the handler is the service that will be invoked
 	Handler []*ServiceInstance `json:"handler"`
 
-	robinIndex uint
+	RobinIndex uint `json:"robin_index"`
 }

pkg/internal/directory/command_mapping.go

@@ -1,67 +1,79 @@
 package directory
 
 import (
+	"context"
+	"git.solsynth.dev/hypernet/nexus/pkg/internal/kv"
 	"git.solsynth.dev/hypernet/nexus/pkg/nex"
+	"github.com/goccy/go-json"
 	"github.com/rs/zerolog/log"
 	"github.com/samber/lo"
-	"strings"
-	"sync"
 )
 
-// In commands, we use the map and the mutex because it is usually read and only sometimes write
+const CommandInfoKvPrefix = "nexus.command/"
-var commandDirectory = make(map[string]*Command)
-var commandDirectoryMutex sync.Mutex
-
-func AddCommand(id, method string, tags []string, handler *ServiceInstance) {
-	commandDirectoryMutex.Lock()
-	defer commandDirectoryMutex.Unlock()
 
+func AddCommand(id, method string, tags []string, handler *ServiceInstance) error {
 	if tags == nil {
 		tags = make([]string, 0)
 	}
 
-	ky := nex.GetCommandKey(id, method)
+	ky := CommandInfoKvPrefix + nex.GetCommandKey(id, method)
-	if _, ok := commandDirectory[ky]; !ok {
+
-		commandDirectory[ky] = &Command{
+	command := &Command{
 		ID:      id,
 		Method:  method,
 		Tags:    tags,
 		Handler: []*ServiceInstance{handler},
 	}
-	} else {
-		commandDirectory[ky].Handler = append(commandDirectory[ky].Handler, handler)
-		commandDirectory[ky].Tags = lo.Uniq(append(commandDirectory[ky].Tags, tags...))
-	}
 
-	commandDirectory[ky].Handler = lo.UniqBy(commandDirectory[ky].Handler, func(item *ServiceInstance) string {
+	command.Handler = lo.UniqBy(command.Handler, func(item *ServiceInstance) string {
 		return item.ID
 	})
 
-	log.Info().Str("id", id).Str("method", method).Str("tags", strings.Join(tags, ",")).Msg("New command registered")
+	commandJSON, err := json.Marshal(command)
+	if err != nil {
+		log.Printf("Error marshaling command: %v", err)
+		return nil
+	}
+
+	_, err = kv.Kv.Put(context.Background(), ky, string(commandJSON))
+	return err
 }
 
 func GetCommandHandler(id, method string) *ServiceInstance {
-	commandDirectoryMutex.Lock()
+	ky := CommandInfoKvPrefix + nex.GetCommandKey(id, method)
-	defer commandDirectoryMutex.Unlock()
 
-	ky := nex.GetCommandKey(id, method)
+	resp, err := kv.Kv.Get(context.Background(), ky)
-	if val, ok := commandDirectory[ky]; ok {
+	if err != nil {
-		if len(val.Handler) == 0 {
 		return nil
 	}
 
-		idx := val.robinIndex % uint(len(val.Handler))
+	if len(resp.Kvs) == 0 {
-		val.robinIndex = idx + 1
+		return nil
-		return val.Handler[idx]
 	}
 
+	var command Command
+	if err := json.Unmarshal(resp.Kvs[0].Value, &command); err != nil {
 		return nil
+	}
+
+	if len(command.Handler) == 0 {
+		return nil
+	}
+
+	idx := command.RobinIndex % uint(len(command.Handler))
+	command.RobinIndex = idx + 1
+
+	raw, err := json.Marshal(&command)
+	if err == nil {
+		_, _ = kv.Kv.Put(context.Background(), ky, string(raw))
+	}
+
+	return command.Handler[idx]
 }
 
-func RemoveCommand(id, method string) {
+func RemoveCommand(id, method string) error {
-	commandDirectoryMutex.Lock()
+	ky := CommandInfoKvPrefix + nex.GetCommandKey(id, method)
-	defer commandDirectoryMutex.Unlock()
 
-	ky := nex.GetCommandKey(id, method)
+	_, err := kv.Kv.Delete(context.Background(), ky)
-	delete(commandDirectory, ky)
+	return err
 }

pkg/internal/directory/service_mapping.go

@@ -2,77 +2,137 @@ package directory
 
 import (
 	"context"
+	"git.solsynth.dev/hypernet/nexus/pkg/internal/kv"
 	"git.solsynth.dev/hypernet/nexus/pkg/nex"
 	"git.solsynth.dev/hypernet/nexus/pkg/proto"
-	"sync"
+	"github.com/goccy/go-json"
+	clientv3 "go.etcd.io/etcd/client/v3"
+	"math/rand"
 	"time"
 )
 
-// In services, we use sync.Map because it will be both often read and write
+const ServiceInfoKvPrefix = "nexus.service/"
-var serviceDirectory sync.Map
 
-func GetServiceInstance(id string) *ServiceInstance {
+func AddServiceInstance(in *ServiceInstance) error {
-	val, ok := serviceDirectory.Load(id)
+	key := ServiceInfoKvPrefix + in.ID
-	if ok {
+	data, err := json.Marshal(in)
-		return val.(*ServiceInstance)
+	if err != nil {
-	} else {
+		return err
-		return nil
 	}
+
+	_, err = kv.Kv.Put(context.Background(), key, string(data))
+	return err
 }
 
-func GetServiceInstanceByType(t string) *ServiceInstance {
+func GetServiceInstance(id string) *ServiceInstance {
-	var result *ServiceInstance
+	key := ServiceInfoKvPrefix + id
-	serviceDirectory.Range(func(key, value any) bool {
+	resp, err := kv.Kv.Get(context.Background(), key)
-		if value.(*ServiceInstance).Type == t {
+	if err != nil || len(resp.Kvs) == 0 {
-			result = value.(*ServiceInstance)
+		return nil
-			return false
 	}
-		return true
+
-	})
+	var instance ServiceInstance
-	return result
+	err = json.Unmarshal(resp.Kvs[0].Value, &instance)
+	if err != nil {
+		return nil
+	}
+
+	return &instance
 }
 
 func ListServiceInstance() []*ServiceInstance {
+	resp, err := kv.Kv.Get(context.Background(), ServiceInfoKvPrefix, clientv3.WithPrefix())
+	if err != nil {
+		return nil
+	}
+
 	var result []*ServiceInstance
-	serviceDirectory.Range(func(key, value interface{}) bool {
+	for _, val := range resp.Kvs {
-		result = append(result, value.(*ServiceInstance))
+		var instance ServiceInstance
-		return true
+		if err := json.Unmarshal(val.Value, &instance); err != nil {
-	})
+			continue
+		}
+		result = append(result, &instance)
+	}
 	return result
 }
 
 func ListServiceInstanceByType(t string) []*ServiceInstance {
-	var result []*ServiceInstance
+	resp, err := kv.Kv.Get(context.Background(), ServiceInfoKvPrefix, clientv3.WithPrefix())
-	serviceDirectory.Range(func(key, value interface{}) bool {
+	if err != nil {
-		if value.(*ServiceInstance).Type == t {
+		return nil
-			result = append(result, value.(*ServiceInstance))
+	}
+
+	var result []*ServiceInstance
+	for _, val := range resp.Kvs {
+		var instance ServiceInstance
+		if err := json.Unmarshal(val.Value, &instance); err != nil {
+			continue
+		}
+		if instance.Type == t {
+			result = append(result, &instance)
+		}
 	}
-		return true
-	})
 	return result
 }
 
-func AddServiceInstance(in *ServiceInstance) {
+var srvRng = rand.New(rand.NewSource(time.Now().UnixNano()))
-	serviceDirectory.Store(in.ID, in)
-}
 
-func RemoveServiceInstance(id string) {
+func GetServiceInstanceByType(t string) *ServiceInstance {
-	serviceDirectory.Delete(id)
+	resp, err := kv.Kv.Get(context.Background(), ServiceInfoKvPrefix, clientv3.WithPrefix())
-}
-
-func BroadcastEvent(event string, data any) {
-	serviceDirectory.Range(func(key, value any) bool {
-		conn, err := value.(*ServiceInstance).GetGrpcConn()
 	if err != nil {
-			return true
+		return nil
+	}
+
+	var instances []*ServiceInstance
+	for _, val := range resp.Kvs {
+		var instance ServiceInstance
+		if err := json.Unmarshal(val.Value, &instance); err != nil {
+			continue
+		}
+		if instance.Type == t {
+			instances = append(instances, &instance)
+		}
+	}
+
+	if len(instances) == 0 {
+		return nil
+	}
+
+	idx := srvRng.Intn(len(instances))
+	return instances[idx]
+}
+
+func RemoveServiceInstance(id string) error {
+	key := ServiceInfoKvPrefix + id
+	_, err := kv.Kv.Delete(context.Background(), key)
+	return err
+}
+
+func BroadcastEvent(event string, data any) error {
+	resp, err := kv.Kv.Get(context.Background(), ServiceInfoKvPrefix, clientv3.WithPrefix())
+	if err != nil {
+		return err
+	}
+
+	for _, val := range resp.Kvs {
+		var instance ServiceInstance
+		if err := json.Unmarshal(val.Value, &instance); err != nil {
+			continue
+		}
+
+		conn, err := instance.GetGrpcConn()
+		if err != nil {
+			continue
 		}
 
 		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-		defer cancel()
 		_, _ = proto.NewDirectoryServiceClient(conn).BroadcastEvent(ctx, &proto.EventInfo{
 			Event: event,
 			Data:  nex.EncodeMap(data),
 		})
-		return true
+		cancel()
-	})
+	}
+
+	return nil
 }

pkg/internal/directory/service_rpc.go

@@ -75,17 +75,25 @@ func (v *ServiceRpcServer) AddService(ctx context.Context, info *proto.ServiceIn
 		GrpcAddr: info.GetGrpcAddr(),
 		HttpAddr: info.HttpAddr,
 	}
-	AddServiceInstance(in)
+	err = AddServiceInstance(in)
+	if err == nil {
 		log.Info().Str("id", clientId).Str("label", info.GetLabel()).Msg("New service registered")
+	} else {
+		log.Error().Str("id", clientId).Str("label", info.GetLabel()).Err(err).Msg("Unable to register a service")
+	}
 	return &proto.AddServiceResponse{
-		IsSuccess: true,
+		IsSuccess: err == nil,
 	}, nil
 }
 
 func (v *ServiceRpcServer) RemoveService(ctx context.Context, request *proto.RemoveServiceRequest) (*proto.RemoveServiceResponse, error) {
-	RemoveServiceInstance(request.GetId())
+	err := RemoveServiceInstance(request.GetId())
-	log.Info().Str("id", request.GetId()).Msg("A service removed.")
+	if err == nil {
+		log.Info().Str("id", request.GetId()).Msg("A service removed")
+	} else {
+		log.Error().Str("id", request.GetId()).Err(err).Msg("Unable to remove a service")
+	}
 	return &proto.RemoveServiceResponse{
-		IsSuccess: true,
+		IsSuccess: err == nil,
 	}, nil
 }

pkg/internal/http/api/directory.go

@@ -1,15 +1,15 @@
 package api
 
 import (
-	directory2 "git.solsynth.dev/hypernet/nexus/pkg/internal/directory"
+	"git.solsynth.dev/hypernet/nexus/pkg/internal/directory"
 	"github.com/gofiber/fiber/v2"
 	"github.com/samber/lo"
 )
 
 func listExistsService(c *fiber.Ctx) error {
-	services := directory2.ListServiceInstance()
+	services := directory.ListServiceInstance()
 
-	return c.JSON(lo.Map(services, func(item *directory2.ServiceInstance, index int) map[string]any {
+	return c.JSON(lo.Map(services, func(item *directory.ServiceInstance, index int) map[string]any {
 		return map[string]any{
 			"id":    item.ID,
 			"type":  item.Type,

pkg/internal/kv/etcd.go

@@ -1,6 +1,10 @@
 package kv
 
 import (
+	"context"
+	"fmt"
+	"github.com/rs/zerolog/log"
+	"github.com/samber/lo"
 	clientv3 "go.etcd.io/etcd/client/v3"
 	"time"
 )
@@ -12,8 +16,22 @@ func ConnectEtcd(endpoints []string) error {
 		Endpoints:   endpoints,
 		DialTimeout: 10 * time.Second,
 	})
-	if err == nil {
+	if err != nil {
-		Kv = conn
+		return err
 	}
+	var status []bool
+	for _, endpoint := range endpoints {
+		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+		_, err := conn.Status(ctx, endpoint)
+		if err != nil {
+			log.Warn().Str("endpoint", endpoint).Err(err).Msg("An KV endpoint is not available...")
+		}
+		status = append(status, err == nil)
+		cancel()
+	}
+	if len(lo.Filter(status, func(s bool, _ int) bool { return s })) == 0 {
+		return fmt.Errorf("unable to connect to all KV endpoints")
+	}
+	Kv = conn
 	return err
 }

pkg/main.go

@@ -47,12 +47,16 @@ func main() {
 	}
 
 	// Connect to kv (etcd)
+	log.Info().Msg("Connecting to kv (etcd)...")
 	if err := kv.ConnectEtcd(viper.GetStringSlice("kv.endpoints")); err != nil {
 		log.Error().Err(err).Msg("An error occurred when connecting to kv (etcd), please check your configuration in kv section.")
 		log.Fatal().Msg("Kv is required for service discovery and directory feature, cannot be disabled.")
+	} else {
+		log.Info().Msg("Connected to kv (etcd)!")
 	}
 
 	// Connect to database
+	log.Info().Msg("Connecting to database...")
 	if db, err := database.Connect(viper.GetString("database.dsn")); err != nil {
 		log.Error().Err(err).Msg("An error occurred when connecting to database. Database related features will be disabled.")
 	} else {
@@ -62,7 +66,7 @@ func main() {
 			log.Error().Err(err).Msg("An error occurred when querying database version. Database related features will be disabled.")
 			database.Kdb = nil
 		} else {
-			log.Info().Str("version", version).Msg("Connected to database")
+			log.Info().Str("version", version).Msg("Connected to database!")
 		}
 	}