优化结构

2026-08-22 08:25:25 +08:00 · 2026-08-22 08:25:25 +08:00 · 2b78b195c1
parent 626f505b5d
commit 2b78b195c1
10 changed files with 815 additions and 115 deletions
--- a/internal/biz/integration_config.go
+++ b/internal/biz/integration_config.go
@ -233,6 +233,9 @@ func validateCommunicationIntegrationConfig(kind, provider string, values map[st
 		if timeout := integrationInt64(values, "connect_timeout", 0); timeout <= 0 {
 			return errors.New("emqx connect_timeout 必须大于 0")
 		}
+		if interval := integrationInt64(values, "reconnect_interval", 0); interval <= 0 {
+			return errors.New("emqx reconnect_interval 必须大于 0")
+		}
 	case IntegrationKindMQ + "/rabbitmq":
 		port := integrationInt64(values, "port", 0)
 		if port < 1 || port > 65535 {
@ -251,6 +254,9 @@ func validateCommunicationIntegrationConfig(kind, provider string, values map[st
 		if integrationInt64(values, "connect_timeout", 0) <= 0 {
 			return errors.New("rabbitmq connect_timeout 必须大于 0")
 		}
+		if integrationInt64(values, "reconnect_interval", 0) <= 0 {
+			return errors.New("rabbitmq reconnect_interval 必须大于 0")
+		}
 	case IntegrationKindWebSocket + "/melody":
 		path := integrationText(values, "path")
 		if !strings.HasPrefix(path, "/") {
--- a/internal/biz/integration_config_definition.go
+++ b/internal/biz/integration_config_definition.go
@ -92,7 +92,7 @@ var integrationDefinitions = map[string][]IntegrationConfigDefinition{
 	IntegrationKindMQ: {
 		{
 			Kind: IntegrationKindMQ, Provider: "emqx", Name: "EMQX", Description: "EMQX MQTT 消息服务",
-			Defaults: map[string]any{"broker": "tcp://127.0.0.1:1883", "client_id": "kra", "username": "", "password": "", "keep_alive": 30, "clean_session": true, "connect_timeout": 10},
+			Defaults: map[string]any{"broker": "tcp://127.0.0.1:1883", "client_id": "kra", "username": "", "password": "", "keep_alive": 30, "clean_session": true, "connect_timeout": 10, "reconnect_interval": 5},
 			Fields: []IntegrationConfigField{
 				{Key: "broker", Label: "Broker 地址", Type: "text", Required: true, Placeholder: "tcp://127.0.0.1:1883"},
 				{Key: "client_id", Label: "客户端 ID", Type: "text", Required: true, Placeholder: "kra"},
@ -101,11 +101,12 @@ var integrationDefinitions = map[string][]IntegrationConfigDefinition{
 				{Key: "keep_alive", Label: "心跳间隔（秒）", Type: "number", Required: true},
 				{Key: "clean_session", Label: "清理会话", Type: "switch", Description: "连接时不恢复 Broker 端保存的旧会话。"},
 				{Key: "connect_timeout", Label: "连接超时（秒）", Type: "number", Required: true},
+				{Key: "reconnect_interval", Label: "重连退避上限（秒）", Type: "number", Required: true, Description: "网络中断后自动重连的最大退避间隔。"},
 			},
 		},
 		{
 			Kind: IntegrationKindMQ, Provider: "rabbitmq", Name: "RabbitMQ", Description: "RabbitMQ AMQP 消息队列",
-			Defaults: map[string]any{"host": "127.0.0.1", "port": 5672, "username": "guest", "password": "guest", "vhost": "/", "exchange": "kra", "exchange_type": "topic", "queue": "kra", "routing_key": "#", "durable": true, "auto_delete": false, "prefetch_count": 10, "heartbeat": 10, "connect_timeout": 10, "tls": false},
+			Defaults: map[string]any{"host": "127.0.0.1", "port": 5672, "username": "guest", "password": "guest", "vhost": "/", "exchange": "kra", "exchange_type": "topic", "queue": "kra", "routing_key": "#", "durable": true, "auto_delete": false, "prefetch_count": 10, "heartbeat": 10, "connect_timeout": 10, "reconnect_interval": 5, "tls": false},
 			Fields: []IntegrationConfigField{
 				{Key: "host", Label: "主机", Type: "text", Required: true, Placeholder: "127.0.0.1"},
 				{Key: "port", Label: "端口", Type: "number", Required: true},
@ -121,6 +122,7 @@ var integrationDefinitions = map[string][]IntegrationConfigDefinition{
 				{Key: "prefetch_count", Label: "预取数量", Type: "number"},
 				{Key: "heartbeat", Label: "心跳间隔（秒）", Type: "number"},
 				{Key: "connect_timeout", Label: "连接超时（秒）", Type: "number", Required: true},
+				{Key: "reconnect_interval", Label: "重连退避上限（秒）", Type: "number", Required: true, Description: "网络中断后自动重连的最大退避间隔。"},
 				{Key: "tls", Label: "启用 TLS", Type: "switch"},
 			},
 		},
--- a/internal/integration/mq/emqx.go
+++ b/internal/integration/mq/emqx.go
@ -20,8 +20,8 @@ const (
 	retryTick        = time.Second
 )

-// Reloadable owns the process-wide message clients. Configuration comes only
-// from sys_integration_configs through runtimeconfig.Store.
+// Reloadable owns process-wide message clients and the logical subscription
+// declarations used to restore them after reconnects or configuration reloads.
 type Reloadable struct {
 	mu            sync.RWMutex
 	opMu          sync.Mutex
@ -31,10 +31,10 @@ type Reloadable struct {
 	bindings      map[string]map[string]byte
 	pending       map[string]bool
 	nextRetry     map[string]time.Time
+	legacySeq     uint64
 	stop          []func()
 	retryStop     chan struct{}
 	retryDone     chan struct{}
-	closeOnce     sync.Once
 	logger        *slog.Logger
 	closed        bool
 }
@ -139,21 +139,24 @@ func (r *Reloadable) apply(provider string, config runtimeconfig.Config) {
 	if r.closed {
 		return
 	}
+	r.ensureStateLocked()
+	config.Kind = "mq"
 	config.Provider = provider
 	config.Values = append(json.RawMessage(nil), config.Values...)
 	r.configs[provider] = config
+	r.replaceClientLocked(provider, nil)
 	if !config.Enabled {
 		delete(r.pending, provider)
 		delete(r.nextRetry, provider)
-		r.replaceClientLocked(provider, nil)
 		return
 	}
 	if err := r.activateLocked(provider, config); err != nil {
-		r.pending[provider] = true
-		r.nextRetry[provider] = time.Now().Add(configRetryInterval(config.Values))
+		r.scheduleRetryLocked(provider, config.Values)
+		if r.logger != nil {
 			r.logger.Warn("message integration unavailable", "mod", "mq", "provider", provider, "error", err)
 		}
 	}
+}

 func (r *Reloadable) activateLocked(provider string, config runtimeconfig.Config) error {
 	client, err := newProviderClient(provider, config.Values)
@ -167,6 +170,9 @@ func (r *Reloadable) activateLocked(provider string, config runtimeconfig.Config
 	}
 	r.replaceClientLocked(provider, client)
 	r.mu.Lock()
+	if r.bindings == nil {
+		r.bindings = make(map[string]map[string]byte)
+	}
 	r.bindings[provider] = bindings
 	r.mu.Unlock()
 	delete(r.pending, provider)
@ -253,8 +259,89 @@ func configBool(values map[string]any, key string) bool {
 	return value
 }

+func (r *Reloadable) retryLoop() {
+	defer close(r.retryDone)
+	ticker := time.NewTicker(retryTick)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-r.retryStop:
+			return
+		case <-ticker.C:
+			r.retryOnce()
+		}
+	}
+}
+
+func (r *Reloadable) retryOnce() {
+	r.opMu.Lock()
+	defer r.opMu.Unlock()
+	if r.closed {
+		return
+	}
+	r.ensureStateLocked()
+	now := time.Now()
+	for _, provider := range []string{ProviderEMQX, ProviderRabbitMQ} {
+		config, exists := r.configs[provider]
+		if !exists || !config.Enabled {
+			continue
+		}
+		client := r.clientLocked(provider)
+		if client != nil && client.Connected() {
+			if r.pending[provider] {
+				if err := r.reconcileProviderLocked(provider); err != nil {
+					r.scheduleRetryLocked(provider, config.Values)
+				}
+			}
+			continue
+		}
+		if client != nil {
+			if recovering, ok := client.(interface{ Reconnecting() bool }); ok && recovering.Reconnecting() {
+				continue
+			}
+		}
+		if retryAt := r.nextRetry[provider]; !retryAt.IsZero() && now.Before(retryAt) {
+			continue
+		}
+		r.replaceClientLocked(provider, nil)
+		if err := r.activateLocked(provider, config); err != nil {
+			r.scheduleRetryLocked(provider, config.Values)
+			if r.logger != nil {
+				r.logger.Warn("message integration reconnect failed", "mod", "mq", "provider", provider, "error", err)
+			}
+		}
+	}
+}
+
+func (r *Reloadable) scheduleRetryLocked(provider string, raw json.RawMessage) {
+	if r.pending == nil {
+		r.pending = make(map[string]bool)
+	}
+	if r.nextRetry == nil {
+		r.nextRetry = make(map[string]time.Time)
+	}
+	r.pending[provider] = true
+	r.nextRetry[provider] = time.Now().Add(configRetryInterval(raw))
+}
+
+func configRetryInterval(raw json.RawMessage) time.Duration {
+	values := map[string]any{}
+	_ = json.Unmarshal(raw, &values)
+	interval := configSeconds(values, "reconnect_interval")
+	if interval <= 0 {
+		return 5 * time.Second
+	}
+	return interval
+}
+
 func (r *Reloadable) replaceClientLocked(provider string, next platformmq.Client) {
 	r.mu.Lock()
+	if r.clients == nil {
+		r.clients = make(map[string]platformmq.Client)
+	}
+	if r.bindings == nil {
+		r.bindings = make(map[string]map[string]byte)
+	}
 	old := r.clients[provider]
 	if next == nil {
 		delete(r.clients, provider)
@ -263,7 +350,7 @@ func (r *Reloadable) replaceClientLocked(provider string, next platformmq.Client
 		r.clients[provider] = next
 	}
 	r.mu.Unlock()
-	if old != nil {
+	if old != nil && old != next {
 		_ = old.Close()
 	}
 }
@ -280,23 +367,241 @@ func (r *Reloadable) restoreSubscriptionsLocked(provider string, client platform
 	return bindings, nil
 }

-func configRetryInterval(raw json.RawMessage) time.Duration {
-	values := map[string]any{}
-	_ = json.Unmarshal(raw, &values)
-	interval := configSeconds(values, "reconnect_interval")
-	if interval <= 0 {
-		return 5 * time.Second
+func (r *Reloadable) desiredSubscriptions(provider string) map[string]byte {
+	result := make(map[string]byte)
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	for topic, owners := range r.subscriptions[provider] {
+		for _, item := range owners {
+			if qos, exists := result[topic]; !exists || item.qos > qos {
+				result[topic] = item.qos
 			}
-	return interval
+		}
+	}
+	return result
 }

-func (r *Reloadable) client(provider string) platformmq.Client {
-	provider = strings.ToLower(strings.TrimSpace(provider))
+func (r *Reloadable) dispatcher(provider, topic string) platformmq.Handler {
+	return func(ctx context.Context, message platformmq.Message) {
+		r.mu.RLock()
+		owners := r.subscriptions[provider][topic]
+		handlers := make([]platformmq.Handler, 0, len(owners))
+		for _, item := range owners {
+			handlers = append(handlers, item.handler)
+		}
+		r.mu.RUnlock()
+		for _, handler := range handlers {
+			handler(ctx, message)
+		}
+	}
+}
+
+func (r *Reloadable) clientLocked(provider string) platformmq.Client {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
 	return r.clients[provider]
 }

+func (r *Reloadable) reconcileProviderLocked(provider string) error {
+	client := r.clientLocked(provider)
+	desired := r.desiredSubscriptions(provider)
+	if client == nil || !client.Connected() {
+		if len(desired) == 0 {
+			delete(r.pending, provider)
+			delete(r.nextRetry, provider)
+			return nil
+		}
+		return platformmq.ErrUnavailable
+	}
+	r.mu.RLock()
+	current := make(map[string]byte, len(r.bindings[provider]))
+	for topic, qos := range r.bindings[provider] {
+		current[topic] = qos
+	}
+	r.mu.RUnlock()
+	for topic := range current {
+		if _, exists := desired[topic]; exists {
+			continue
+		}
+		if err := client.Unsubscribe(context.Background(), topic); err != nil {
+			return err
+		}
+		delete(current, topic)
+	}
+	for topic, qos := range desired {
+		if oldQoS, exists := current[topic]; exists && oldQoS == qos {
+			continue
+		}
+		if _, exists := current[topic]; exists {
+			if err := client.Unsubscribe(context.Background(), topic); err != nil {
+				return err
+			}
+		}
+		if err := client.Subscribe(context.Background(), topic, qos, r.dispatcher(provider, topic)); err != nil {
+			return err
+		}
+		current[topic] = qos
+	}
+	r.mu.Lock()
+	r.bindings[provider] = current
+	r.mu.Unlock()
+	delete(r.pending, provider)
+	delete(r.nextRetry, provider)
+	return nil
+}
+
+func (r *Reloadable) Register(set platformmq.SubscriptionSet) error {
+	normalized, err := platformmq.NormalizeSubscriptionSet(set)
+	if err != nil {
+		return err
+	}
+	r.opMu.Lock()
+	defer r.opMu.Unlock()
+	if r.closed {
+		return platformmq.ErrUnavailable
+	}
+	r.ensureStateLocked()
+	r.mu.Lock()
+	if r.subscriptions[normalized.Provider] == nil {
+		r.subscriptions[normalized.Provider] = make(map[string]map[string]subscription)
+	}
+	for topic, owners := range r.subscriptions[normalized.Provider] {
+		delete(owners, normalized.Owner)
+		if len(owners) == 0 {
+			delete(r.subscriptions[normalized.Provider], topic)
+		}
+	}
+	for _, item := range normalized.Topics {
+		if r.subscriptions[normalized.Provider][item.Topic] == nil {
+			r.subscriptions[normalized.Provider][item.Topic] = make(map[string]subscription)
+		}
+		r.subscriptions[normalized.Provider][item.Topic][normalized.Owner] = subscription{qos: item.QoS, handler: item.Handler}
+	}
+	r.mu.Unlock()
+	if err := r.reconcileProviderLocked(normalized.Provider); err != nil {
+		config := r.configs[normalized.Provider]
+		r.scheduleRetryLocked(normalized.Provider, config.Values)
+		if r.logger != nil {
+			r.logger.Warn("message subscription bind deferred", "mod", "mq", "provider", normalized.Provider, "owner", normalized.Owner, "error", err)
+		}
+	}
+	return nil
+}
+
+func (r *Reloadable) Unregister(owner string) error {
+	owner = strings.TrimSpace(owner)
+	if owner == "" {
+		return fmt.Errorf("mq subscription owner is empty")
+	}
+	r.opMu.Lock()
+	defer r.opMu.Unlock()
+	if r.closed {
+		return nil
+	}
+	r.ensureStateLocked()
+	r.mu.Lock()
+	providers := make([]string, 0, len(r.subscriptions))
+	for provider, topics := range r.subscriptions {
+		providers = append(providers, provider)
+		for topic, owners := range topics {
+			delete(owners, owner)
+			if len(owners) == 0 {
+				delete(topics, topic)
+			}
+		}
+	}
+	r.mu.Unlock()
+	for _, provider := range providers {
+		if err := r.reconcileProviderLocked(provider); err != nil {
+			if config, ok := r.configs[provider]; ok {
+				r.scheduleRetryLocked(provider, config.Values)
+			}
+		}
+	}
+	return nil
+}
+
+func (r *Reloadable) Publish(ctx context.Context, topic string, payload []byte, qos byte, retain bool) error {
+	return r.PublishTo(ctx, ProviderEMQX, topic, payload, qos, retain)
+}
+
+func (r *Reloadable) PublishTo(ctx context.Context, provider, topic string, payload []byte, qos byte, retain bool) error {
+	provider = strings.ToLower(strings.TrimSpace(provider))
+	client := r.clientLocked(provider)
+	if client == nil {
+		return platformmq.ErrUnavailable
+	}
+	return client.Publish(ctx, topic, payload, qos, retain)
+}
+
+func (r *Reloadable) Subscribe(ctx context.Context, topic string, qos byte, handler platformmq.Handler) error {
+	return r.SubscribeTo(ctx, ProviderEMQX, topic, qos, handler)
+}
+
+func (r *Reloadable) SubscribeTo(ctx context.Context, provider, topic string, qos byte, handler platformmq.Handler) error {
+	if ctx != nil {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		default:
+		}
+	}
+	provider = strings.ToLower(strings.TrimSpace(provider))
+	r.opMu.Lock()
+	r.legacySeq++
+	owner := fmt.Sprintf("legacy/%s/%d", provider, r.legacySeq)
+	r.opMu.Unlock()
+	if err := r.Register(platformmq.SubscriptionSet{Owner: owner, Provider: provider, Topics: []platformmq.TopicSubscription{{Topic: topic, QoS: qos, Handler: handler}}}); err != nil {
+		return err
+	}
+	if !r.ConnectedTo(provider) {
+		return platformmq.ErrUnavailable
+	}
+	return nil
+}
+
+func (r *Reloadable) Unsubscribe(ctx context.Context, topics ...string) error {
+	return r.UnsubscribeFrom(ctx, ProviderEMQX, topics...)
+}
+
+func (r *Reloadable) UnsubscribeFrom(ctx context.Context, provider string, topics ...string) error {
+	if ctx != nil {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		default:
+		}
+	}
+	provider = strings.ToLower(strings.TrimSpace(provider))
+	if len(topics) == 0 {
+		return fmt.Errorf("mq topics are empty")
+	}
+	r.opMu.Lock()
+	defer r.opMu.Unlock()
+	r.ensureStateLocked()
+	r.mu.Lock()
+	for _, rawTopic := range topics {
+		topic := strings.TrimSpace(rawTopic)
+		owners := r.subscriptions[provider][topic]
+		for owner := range owners {
+			if strings.HasPrefix(owner, "legacy/") {
+				delete(owners, owner)
+			}
+		}
+		if len(owners) == 0 {
+			delete(r.subscriptions[provider], topic)
+		}
+	}
+	r.mu.Unlock()
+	if err := r.reconcileProviderLocked(provider); err != nil {
+		if config, ok := r.configs[provider]; ok {
+			r.scheduleRetryLocked(provider, config.Values)
+		}
+		return err
+	}
+	return nil
+}
+
 func (r *Reloadable) Client(provider string) platformmq.Client {
 	provider = strings.ToLower(strings.TrimSpace(provider))
 	if provider != ProviderEMQX && provider != ProviderRabbitMQ {
@ -317,81 +622,24 @@ func (c *namedClient) Unsubscribe(ctx context.Context, topics ...string) error {
 func (c *namedClient) Connected() bool { return c.owner.ConnectedTo(c.provider) }
 func (*namedClient) Close() error      { return nil }

-func (r *Reloadable) Publish(ctx context.Context, topic string, payload []byte, qos byte, retain bool) error {
-	return r.PublishTo(ctx, ProviderEMQX, topic, payload, qos, retain)
-}
-
-func (r *Reloadable) PublishTo(ctx context.Context, provider, topic string, payload []byte, qos byte, retain bool) error {
-	provider = strings.ToLower(strings.TrimSpace(provider))
-	r.mu.RLock()
-	defer r.mu.RUnlock()
-	client := r.clients[provider]
-	if client == nil {
-		return platformmq.ErrUnavailable
-	}
-	return client.Publish(ctx, topic, payload, qos, retain)
-}
-
-func (r *Reloadable) Subscribe(ctx context.Context, topic string, qos byte, handler platformmq.Handler) error {
-	return r.SubscribeTo(ctx, ProviderEMQX, topic, qos, handler)
-}
-
-func (r *Reloadable) SubscribeTo(ctx context.Context, provider, topic string, qos byte, handler platformmq.Handler) error {
-	provider = strings.ToLower(strings.TrimSpace(provider))
-	r.opMu.Lock()
-	defer r.opMu.Unlock()
-	client := r.client(provider)
-	if client == nil {
-		return platformmq.ErrUnavailable
-	}
-	if err := client.Subscribe(ctx, topic, qos, handler); err != nil {
-		return err
-	}
-	if r.subscriptions[provider] == nil {
-		r.subscriptions[provider] = make(map[string]subscription)
-	}
-	r.subscriptions[provider][topic] = subscription{qos: qos, handler: handler}
-	return nil
-}
-
-func (r *Reloadable) Unsubscribe(ctx context.Context, topics ...string) error {
-	return r.UnsubscribeFrom(ctx, ProviderEMQX, topics...)
-}
-
-func (r *Reloadable) UnsubscribeFrom(ctx context.Context, provider string, topics ...string) error {
-	provider = strings.ToLower(strings.TrimSpace(provider))
-	r.opMu.Lock()
-	defer r.opMu.Unlock()
-	client := r.client(provider)
-	if client == nil {
-		return platformmq.ErrUnavailable
-	}
-	if err := client.Unsubscribe(ctx, topics...); err != nil {
-		return err
-	}
-	for _, topic := range topics {
-		delete(r.subscriptions[provider], topic)
-	}
-	return nil
-}
-
 func (r *Reloadable) Connected() bool { return r.ConnectedTo(ProviderEMQX) }

 func (r *Reloadable) ConnectedTo(provider string) bool {
 	provider = strings.ToLower(strings.TrimSpace(provider))
-	r.mu.RLock()
-	defer r.mu.RUnlock()
-	client := r.clients[provider]
+	client := r.clientLocked(provider)
 	return client != nil && client.Connected()
 }

 func (r *Reloadable) Close() error {
 	r.opMu.Lock()
-	defer r.opMu.Unlock()
 	if r.closed {
+		r.opMu.Unlock()
 		return nil
 	}
 	r.closed = true
+	if r.retryStop != nil {
+		close(r.retryStop)
+	}
 	r.mu.Lock()
 	clients := make([]platformmq.Client, 0, len(r.clients))
 	for provider, client := range r.clients {
@ -399,6 +647,10 @@ func (r *Reloadable) Close() error {
 		delete(r.clients, provider)
 	}
 	r.mu.Unlock()
+	r.opMu.Unlock()
+	if r.retryDone != nil {
+		<-r.retryDone
+	}
 	for _, client := range clients {
 		if client != nil {
 			_ = client.Close()
@ -406,3 +658,24 @@ func (r *Reloadable) Close() error {
 	}
 	return nil
 }
+
+func (r *Reloadable) ensureStateLocked() {
+	if r.clients == nil {
+		r.clients = make(map[string]platformmq.Client)
+	}
+	if r.configs == nil {
+		r.configs = make(map[string]runtimeconfig.Config)
+	}
+	if r.subscriptions == nil {
+		r.subscriptions = make(map[string]map[string]map[string]subscription)
+	}
+	if r.bindings == nil {
+		r.bindings = make(map[string]map[string]byte)
+	}
+	if r.pending == nil {
+		r.pending = make(map[string]bool)
+	}
+	if r.nextRetry == nil {
+		r.nextRetry = make(map[string]time.Time)
+	}
+}
--- a/internal/integration/mq/emqx_test.go
+++ b/internal/integration/mq/emqx_test.go
@ -2,19 +2,27 @@ package mq

 import (
 	"context"
+	"log/slog"
 	"testing"
+	"time"

+	"kra/internal/integration/runtimeconfig"
 	platformmq "kra/pkg/mq"
 )

 type fakeClient struct {
 	subscribed   []string
 	unsubscribed []string
+	handlers     map[string]platformmq.Handler
 }

 func (*fakeClient) Publish(context.Context, string, []byte, byte, bool) error { return nil }
-func (f *fakeClient) Subscribe(_ context.Context, topic string, _ byte, _ platformmq.Handler) error {
+func (f *fakeClient) Subscribe(_ context.Context, topic string, _ byte, handler platformmq.Handler) error {
 	f.subscribed = append(f.subscribed, topic)
+	if f.handlers == nil {
+		f.handlers = make(map[string]platformmq.Handler)
+	}
+	f.handlers[topic] = handler
 	return nil
 }
 func (f *fakeClient) Unsubscribe(_ context.Context, topics ...string) error {
@ -28,7 +36,11 @@ func TestReloadableTracksSubscriptions(t *testing.T) {
 	client := &fakeClient{}
 	r := &Reloadable{
 		clients:       map[string]platformmq.Client{ProviderEMQX: client},
-		subscriptions: make(map[string]map[string]subscription),
+		configs:       map[string]runtimeconfig.Config{ProviderEMQX: {Kind: "mq", Provider: ProviderEMQX, Enabled: true}},
+		subscriptions: make(map[string]map[string]map[string]subscription),
+		bindings:      make(map[string]map[string]byte),
+		pending:       make(map[string]bool),
+		nextRetry:     make(map[string]time.Time),
 	}
 	handler := func(context.Context, platformmq.Message) {}
 	if err := r.Subscribe(context.Background(), "orders/+/paid", platformmq.AtLeastOnce, handler); err != nil {
@ -47,15 +59,92 @@ func TestReloadableTracksSubscriptions(t *testing.T) {

 func TestReloadableRestoresSubscriptions(t *testing.T) {
 	client := &fakeClient{}
-	r := &Reloadable{subscriptions: map[string]map[string]subscription{
+	r := &Reloadable{subscriptions: map[string]map[string]map[string]subscription{
 		ProviderEMQX: {
-			"orders/+/paid": {qos: platformmq.AtLeastOnce, handler: func(context.Context, platformmq.Message) {}},
+			"orders/+/paid": {
+				"orders": {qos: platformmq.AtLeastOnce, handler: func(context.Context, platformmq.Message) {}},
+			},
 		},
 	}}
-	if err := r.restoreSubscriptionsLocked(ProviderEMQX, client); err != nil {
+	bindings, err := r.restoreSubscriptionsLocked(ProviderEMQX, client)
+	if err != nil {
 		t.Fatal(err)
 	}
 	if len(client.subscribed) != 1 || client.subscribed[0] != "orders/+/paid" {
 		t.Fatalf("restored subscriptions = %v", client.subscribed)
 	}
+	if bindings["orders/+/paid"] != platformmq.AtLeastOnce {
+		t.Fatalf("restored binding qos = %d", bindings["orders/+/paid"])
+	}
+}
+
+func TestReloadableRegistersWhileOfflineAndRestoresLater(t *testing.T) {
+	client := &fakeClient{}
+	r := &Reloadable{
+		clients:       make(map[string]platformmq.Client),
+		configs:       map[string]runtimeconfig.Config{ProviderEMQX: {Kind: "mq", Provider: ProviderEMQX, Enabled: true}},
+		subscriptions: make(map[string]map[string]map[string]subscription),
+		bindings:      make(map[string]map[string]byte),
+		pending:       make(map[string]bool),
+		nextRetry:     make(map[string]time.Time),
+		logger:        slog.Default(),
+	}
+	called := make(chan struct{}, 1)
+	err := r.Register(platformmq.SubscriptionSet{
+		Owner:    "orders",
+		Provider: ProviderEMQX,
+		Topics:   []platformmq.TopicSubscription{{Topic: "orders.created", QoS: platformmq.AtLeastOnce, Handler: func(context.Context, platformmq.Message) { called <- struct{}{} }}},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !r.pending[ProviderEMQX] {
+		t.Fatal("offline subscription was not marked pending")
+	}
+	r.clients[ProviderEMQX] = client
+	if err = r.reconcileProviderLocked(ProviderEMQX); err != nil {
+		t.Fatal(err)
+	}
+	if len(client.subscribed) != 1 || client.subscribed[0] != "orders.created" {
+		t.Fatalf("restored subscriptions = %v", client.subscribed)
+	}
+}
+
+func TestReloadableDispatchesSameTopicToMultipleOwners(t *testing.T) {
+	client := &fakeClient{}
+	r := &Reloadable{
+		clients:       map[string]platformmq.Client{ProviderEMQX: client},
+		configs:       map[string]runtimeconfig.Config{ProviderEMQX: {Kind: "mq", Provider: ProviderEMQX, Enabled: true}},
+		subscriptions: make(map[string]map[string]map[string]subscription),
+		bindings:      make(map[string]map[string]byte),
+		pending:       make(map[string]bool),
+		nextRetry:     make(map[string]time.Time),
+		logger:        slog.Default(),
+	}
+	first, second := make(chan struct{}, 1), make(chan struct{}, 1)
+	for _, set := range []platformmq.SubscriptionSet{
+		{Owner: "orders", Provider: ProviderEMQX, Topics: []platformmq.TopicSubscription{{Topic: "events.created", Handler: func(context.Context, platformmq.Message) { first <- struct{}{} }}}},
+		{Owner: "audit", Provider: ProviderEMQX, Topics: []platformmq.TopicSubscription{{Topic: "events.created", Handler: func(context.Context, platformmq.Message) { second <- struct{}{} }}}},
+	} {
+		if err := r.Register(set); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if err := r.reconcileProviderLocked(ProviderEMQX); err != nil {
+		t.Fatalf("explicit reconcile failed: %v", err)
+	}
+	if len(client.subscribed) != 1 {
+		t.Fatalf("broker subscriptions = %v, want one shared topic; desired=%#v bindings=%#v clients=%#v", client.subscribed, r.subscriptions, r.bindings, r.clients)
+	}
+	client.handlers["events.created"](context.Background(), platformmq.Message{Topic: "events.created"})
+	select {
+	case <-first:
+	default:
+		t.Fatal("orders handler was not called")
+	}
+	select {
+	case <-second:
+	default:
+		t.Fatal("audit handler was not called")
+	}
 }
--- a/internal/integration/provider.go
+++ b/internal/integration/provider.go
@ -25,6 +25,7 @@ var ProviderSet = wire.NewSet(
 	mqintegration.New,
 	wire.Bind(new(mq.Client), new(*mqintegration.Reloadable)),
 	wire.Bind(new(mq.Registry), new(*mqintegration.Reloadable)),
+	wire.Bind(new(mq.SubscriptionRegistrar), new(*mqintegration.Reloadable)),
 	websocketintegration.New,
 	wire.Bind(new(platformws.Hub), new(*websocketintegration.Server)),
 	wire.Bind(new(biz.FileStorage), new(*storage.Reloadable)),
--- a/pkg/mq/mq.go
+++ b/pkg/mq/mq.go
@ -91,6 +91,7 @@ type Config struct {
 	KeepAlive         time.Duration
 	CleanSession      bool
 	ConnectTimeout    time.Duration
+	ReconnectInterval time.Duration
 }

 func PublishJSON(ctx context.Context, client Client, topic string, value any, qos byte, retain bool) error {
--- a/pkg/mq/mqtt.go
+++ b/pkg/mq/mqtt.go
@ -28,9 +28,12 @@ func NewMQTT(cfg Config) (*MQTT, error) {
 	if cfg.ConnectTimeout <= 0 {
 		cfg.ConnectTimeout = 10 * time.Second
 	}
+	if cfg.ReconnectInterval <= 0 {
+		cfg.ReconnectInterval = 5 * time.Second
+	}
 	opts := paho.NewClientOptions().AddBroker(cfg.Broker).SetClientID(cfg.ClientID).SetUsername(cfg.Username).SetPassword(cfg.Password)
 	opts.SetKeepAlive(cfg.KeepAlive).SetCleanSession(cfg.CleanSession).SetConnectTimeout(cfg.ConnectTimeout).SetAutoReconnect(true)
-	opts.SetResumeSubs(true).SetOrderMatters(false)
+	opts.SetMaxReconnectInterval(cfg.ReconnectInterval).SetResumeSubs(true).SetOrderMatters(false)
 	c := &MQTT{}
 	token := paho.NewClient(opts)
 	connect := token.Connect()
--- a/pkg/mq/rabbitmq.go
+++ b/pkg/mq/rabbitmq.go
@ -6,6 +6,7 @@ import (
 	"fmt"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"

 	amqp "github.com/rabbitmq/amqp091-go"
@ -27,6 +28,7 @@ type RabbitMQConfig struct {
 	PrefetchCount     int
 	Heartbeat         time.Duration
 	ConnectTimeout    time.Duration
+	ReconnectInterval time.Duration
 	TLS               bool
 }

@ -52,6 +54,7 @@ type RabbitMQ struct {
 	consuming      bool
 	stop           chan struct{}
 	closed         bool
+	reconnecting   atomic.Bool
 }

 func NewRabbitMQ(config RabbitMQConfig) (*RabbitMQ, error) {
@ -75,7 +78,13 @@ func NewRabbitMQ(config RabbitMQConfig) (*RabbitMQ, error) {
 	}.String()
 	connection, err := amqp.DialConfig(address, amqp.Config{
 		Heartbeat: config.Heartbeat,
-		Recovery:  &amqp.Recovery{},
+		Recovery: &amqp.Recovery{
+			ReconnectionConfig: &amqp.ReconnectionConfig{MaxRetryCount: 5, RetryInterval: config.ReconnectInterval},
+			OnTopologyEntityError: func(_ *amqp.Connection, entity amqp.TopologyRecoveryEntity) bool {
+				amqp.Logger.Printf("rabbitmq topology recovery failed: %s", entity.Error())
+				return false
+			},
+		},
 	})
 	if err != nil {
 		return nil, fmt.Errorf("connect rabbitmq: %w", err)
@ -116,6 +125,9 @@ func NewRabbitMQ(config RabbitMQConfig) (*RabbitMQ, error) {
 	client.consumeChannel = consumeChannel
 	client.config = config
 	client.consumerTag = fmt.Sprintf("kra-%d", time.Now().UnixNano())
+	stateChanges := make(chan *amqp.StateChanged, 16)
+	connection.NotifyStateChange(stateChanges)
+	go client.watchState(stateChanges)
 	return client, nil
 }

@ -148,6 +160,9 @@ func defaultRabbitMQConfig(config RabbitMQConfig) RabbitMQConfig {
 	if config.ConnectTimeout <= 0 {
 		config.ConnectTimeout = 10 * time.Second
 	}
+	if config.ReconnectInterval <= 0 {
+		config.ReconnectInterval = 5 * time.Second
+	}
 	return config
 }

@ -371,7 +386,28 @@ func (c *RabbitMQ) Connected() bool {
 	}
 	c.mu.RLock()
 	defer c.mu.RUnlock()
-	return !c.closed && c.connection != nil && !c.connection.IsClosed()
+	return !c.closed && !c.reconnecting.Load() && c.connection != nil && !c.connection.IsClosed()
+}
+
+// Reconnecting reports whether the AMQP driver is actively recovering the
+// current connection. The outer runtime waits for this state to settle before
+// deciding whether a fresh client must be built.
+func (c *RabbitMQ) Reconnecting() bool {
+	return c != nil && !c.closed && c.reconnecting.Load()
+}
+
+func (c *RabbitMQ) watchState(states <-chan *amqp.StateChanged) {
+	for state := range states {
+		if state == nil {
+			continue
+		}
+		switch state.To {
+		case amqp.StateReconnecting:
+			c.reconnecting.Store(true)
+		case amqp.StateOpen, amqp.StateClosed:
+			c.reconnecting.Store(false)
+		}
+	}
 }

 func (c *RabbitMQ) Close() error {
--- a/pkg/mq/subscription_test.go
+++ b/pkg/mq/subscription_test.go
@ -0,0 +1,232 @@
+package mq
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"strings"
+	"testing"
+)
+
+func TestNormalizeSubscriptionSetValidDeclaration(t *testing.T) {
+	handler := func(context.Context, Message) {}
+	set := SubscriptionSet{
+		Owner:    "  orders  ",
+		Provider: "  RABBITMQ ",
+		Topics: []TopicSubscription{
+			{Topic: "  orders.created  ", QoS: AtMostOnce, Handler: handler},
+			{Topic: "orders.updated", QoS: AtLeastOnce, Handler: handler},
+			{Topic: "orders.deleted", QoS: ExactlyOnce, Handler: handler},
+		},
+	}
+
+	got, err := NormalizeSubscriptionSet(set)
+	if err != nil {
+		t.Fatalf("NormalizeSubscriptionSet() error = %v", err)
+	}
+
+	want := SubscriptionSet{
+		Owner:    "orders",
+		Provider: ProviderRabbitMQ,
+		Topics: []TopicSubscription{
+			{Topic: "orders.created", QoS: AtMostOnce, Handler: handler},
+			{Topic: "orders.updated", QoS: AtLeastOnce, Handler: handler},
+			{Topic: "orders.deleted", QoS: ExactlyOnce, Handler: handler},
+		},
+	}
+	if got.Owner != want.Owner || got.Provider != want.Provider {
+		t.Fatalf("normalized identity = %#v, want owner=%q provider=%q", got, want.Owner, want.Provider)
+	}
+	if len(got.Topics) != len(want.Topics) {
+		t.Fatalf("normalized topic count = %d, want %d", len(got.Topics), len(want.Topics))
+	}
+	for index := range want.Topics {
+		if got.Topics[index].Topic != want.Topics[index].Topic {
+			t.Errorf("topic[%d] = %q, want %q", index, got.Topics[index].Topic, want.Topics[index].Topic)
+		}
+		if got.Topics[index].QoS != want.Topics[index].QoS {
+			t.Errorf("qos[%d] = %d, want %d", index, got.Topics[index].QoS, want.Topics[index].QoS)
+		}
+		if reflect.ValueOf(got.Topics[index].Handler).Pointer() != reflect.ValueOf(want.Topics[index].Handler).Pointer() {
+			t.Errorf("handler[%d] was changed", index)
+		}
+	}
+}
+
+func TestNormalizeSubscriptionSetRejectsInvalidDeclarations(t *testing.T) {
+	handler := func(context.Context, Message) {}
+	base := func() SubscriptionSet {
+		return SubscriptionSet{
+			Owner:    "orders",
+			Provider: ProviderEMQX,
+			Topics: []TopicSubscription{
+				{Topic: "orders.created", QoS: AtLeastOnce, Handler: handler},
+			},
+		}
+	}
+
+	tests := []struct {
+		name    string
+		set     SubscriptionSet
+		wantErr string
+	}{
+		{
+			name:    "empty owner",
+			set:     func() SubscriptionSet { set := base(); set.Owner = "   "; return set }(),
+			wantErr: "owner is empty",
+		},
+		{
+			name:    "empty provider",
+			set:     func() SubscriptionSet { set := base(); set.Provider = "   "; return set }(),
+			wantErr: "unsupported mq provider",
+		},
+		{
+			name:    "unsupported provider",
+			set:     func() SubscriptionSet { set := base(); set.Provider = "kafka"; return set }(),
+			wantErr: "unsupported mq provider",
+		},
+		{
+			name:    "empty topics",
+			set:     func() SubscriptionSet { set := base(); set.Topics = nil; return set }(),
+			wantErr: "topics are empty",
+		},
+		{
+			name: "empty topic",
+			set: func() SubscriptionSet {
+				set := base()
+				set.Topics[0].Topic = "  "
+				return set
+			}(),
+			wantErr: "topic at index 0 is empty",
+		},
+		{
+			name: "duplicate topic",
+			set: func() SubscriptionSet {
+				set := base()
+				set.Topics = append(set.Topics, TopicSubscription{
+					Topic: " orders.created ", QoS: AtMostOnce, Handler: handler,
+				})
+				return set
+			}(),
+			wantErr: "duplicate mq subscription topic",
+		},
+		{
+			name: "nil handler",
+			set: func() SubscriptionSet {
+				set := base()
+				set.Topics[0].Handler = nil
+				return set
+			}(),
+			wantErr: "handler is nil",
+		},
+		{
+			name: "invalid qos",
+			set: func() SubscriptionSet {
+				set := base()
+				set.Topics[0].QoS = ExactlyOnce + 1
+				return set
+			}(),
+			wantErr: "invalid mq qos",
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			_, err := NormalizeSubscriptionSet(test.set)
+			if err == nil {
+				t.Fatal("NormalizeSubscriptionSet() error = nil, want error")
+			}
+			if !strings.Contains(err.Error(), test.wantErr) {
+				t.Fatalf("error = %q, want substring %q", err, test.wantErr)
+			}
+		})
+	}
+}
+
+type subscriptionTestRegistrar struct {
+	sets []SubscriptionSet
+}
+
+func (r *subscriptionTestRegistrar) Register(set SubscriptionSet) error {
+	r.sets = append(r.sets, set)
+	return nil
+}
+
+func (*subscriptionTestRegistrar) Unregister(string) error { return nil }
+
+type subscriptionTestContributor struct {
+	name  string
+	set   SubscriptionSet
+	order *[]string
+	err   error
+}
+
+func (c *subscriptionTestContributor) RegisterSubscriptions(registrar SubscriptionRegistrar) error {
+	*c.order = append(*c.order, c.name)
+	if c.err != nil {
+		return c.err
+	}
+	return registrar.Register(c.set)
+}
+
+func TestApplySubscriptionsAppliesMultipleContributorsInOrder(t *testing.T) {
+	order := make([]string, 0, 2)
+	registrar := &subscriptionTestRegistrar{}
+	handler := func(context.Context, Message) {}
+	first := &subscriptionTestContributor{
+		name:  "orders",
+		order: &order,
+		set: SubscriptionSet{
+			Owner:    "orders",
+			Provider: ProviderRabbitMQ,
+			Topics:   []TopicSubscription{{Topic: "orders.created", Handler: handler}},
+		},
+	}
+	second := &subscriptionTestContributor{
+		name:  "notifications",
+		order: &order,
+		set: SubscriptionSet{
+			Owner:    "notifications",
+			Provider: ProviderEMQX,
+			Topics:   []TopicSubscription{{Topic: "notifications.sent", Handler: handler}},
+		},
+	}
+
+	if err := ApplySubscriptions(registrar, first, nil, second); err != nil {
+		t.Fatalf("ApplySubscriptions() error = %v", err)
+	}
+	if !reflect.DeepEqual(order, []string{"orders", "notifications"}) {
+		t.Fatalf("contributor order = %#v, want %#v", order, []string{"orders", "notifications"})
+	}
+	if len(registrar.sets) != 2 {
+		t.Fatalf("registered sets = %d, want 2", len(registrar.sets))
+	}
+	if registrar.sets[0].Owner != "orders" || registrar.sets[1].Owner != "notifications" {
+		t.Fatalf("registered owners = %q, %q", registrar.sets[0].Owner, registrar.sets[1].Owner)
+	}
+}
+
+func TestApplySubscriptionsPropagatesContributorError(t *testing.T) {
+	order := make([]string, 0, 2)
+	registrar := &subscriptionTestRegistrar{}
+	wantErr := errors.New("registration failed")
+	failing := &subscriptionTestContributor{name: "failing", order: &order, err: wantErr}
+	following := &subscriptionTestContributor{name: "following", order: &order}
+
+	err := ApplySubscriptions(registrar, failing, following)
+	if !errors.Is(err, wantErr) {
+		t.Fatalf("ApplySubscriptions() error = %v, want %v", err, wantErr)
+	}
+	if !reflect.DeepEqual(order, []string{"failing"}) {
+		t.Fatalf("contributors called = %#v, want %#v", order, []string{"failing"})
+	}
+	if len(registrar.sets) != 0 {
+		t.Fatalf("registered sets = %d, want 0", len(registrar.sets))
+	}
+}
+
+func TestApplySubscriptionsRejectsNilRegistrar(t *testing.T) {
+	if err := ApplySubscriptions(nil); err == nil || !strings.Contains(err.Error(), "registrar is nil") {
+		t.Fatalf("ApplySubscriptions(nil) error = %v, want nil-registrar error", err)
+	}
+}
--- a/web/src/view/systemTools/integration/config.vue
+++ b/web/src/view/systemTools/integration/config.vue
@ -116,6 +116,7 @@
                class="field-control"
                :min="numberConstraint(field.key).min"
                :max="numberConstraint(field.key).max"
+                :precision="numberConstraint(field.key).integer ? 0 : undefined"
                :step="1"
                controls-position="right"
                @update:model-value="clearFieldError(selected, field.key)"
@ -231,10 +232,21 @@ const TARGETS = {
 }

 const TARGET_ORDER = Object.keys(TARGETS)
+const RECONNECT_INTERVAL_KEY = 'reconnect_interval'
+const RECONNECT_DEFAULT_SECONDS = 5
+const MQ_RECONNECT_TARGETS = new Set(['mq/emqx', 'mq/rabbitmq'])
+const RECONNECT_INTERVAL_FIELD = {
+  key: RECONNECT_INTERVAL_KEY,
+  label: '重连间隔（秒）',
+  type: 'number',
+  required: true,
+  description: '连接中断后再次尝试连接的等待时间，最小 1 秒。'
+}
 const NUMBER_CONSTRAINTS = {
  port: { min: 1, max: 65535 },
  keep_alive: { min: 1 },
  connect_timeout: { min: 1 },
+  reconnect_interval: { min: 1, integer: true },
  prefetch_count: { min: 0 },
  heartbeat: { min: 0 },
  max_message_size: { min: 0 },
@ -264,13 +276,56 @@ const errorKey = (item, fieldKey) => `${integrationKey(item)}:${fieldKey}`
 const listKey = (item, fieldKey) => `${integrationKey(item)}:${fieldKey}`
 const cloneConfig = (value) => JSON.parse(JSON.stringify(value || {}))

+const communicationFields = (item) => {
+  const fields = Array.isArray(item.fields)
+    ? item.fields.map((field) =>
+        field.key === RECONNECT_INTERVAL_KEY
+          ? {
+              ...field,
+              label: field.label || RECONNECT_INTERVAL_FIELD.label,
+              type: 'number',
+              required: true,
+              description: field.description || RECONNECT_INTERVAL_FIELD.description
+            }
+          : { ...field }
+      )
+    : []
+  if (
+    !MQ_RECONNECT_TARGETS.has(integrationKey(item)) ||
+    fields.some((field) => field.key === RECONNECT_INTERVAL_KEY)
+  ) {
+    return fields
+  }
+
+  const connectTimeoutIndex = fields.findIndex(
+    (field) => field.key === 'connect_timeout'
+  )
+  fields.splice(
+    connectTimeoutIndex < 0 ? fields.length : connectTimeoutIndex + 1,
+    0,
+    { ...RECONNECT_INTERVAL_FIELD }
+  )
+  return fields
+}
+
 const normalizeIntegration = (item) => {
+  const config = cloneConfig(item.config)
+  const fields = communicationFields(item)
+  if (
+    MQ_RECONNECT_TARGETS.has(integrationKey(item)) &&
+    (config[RECONNECT_INTERVAL_KEY] === null ||
+      typeof config[RECONNECT_INTERVAL_KEY] === 'undefined' ||
+      config[RECONNECT_INTERVAL_KEY] === '')
+  ) {
+    config[RECONNECT_INTERVAL_KEY] = RECONNECT_DEFAULT_SECONDS
+  }
+
  const normalized = {
    ...item,
    enabled: Boolean(item.enabled),
    configured: Boolean(item.configured),
-    config: cloneConfig(item.config),
-    fields: Array.isArray(item.fields) ? item.fields : []
+    config,
+    fields
  }
  normalized._savedEnabled = normalized.enabled
  normalized._savedConfig = cloneConfig(normalized.config)
@ -359,6 +414,8 @@ const validate = (item, enabled = item.enabled) => {
      const constraint = numberConstraint(field.key)
      if (!Number.isFinite(number)) {
        message = `${field.label}必须是数字`
+      } else if (constraint.integer && !Number.isInteger(number)) {
+        message = `${field.label}必须是整数`
      } else if (constraint.min !== undefined && number < constraint.min) {
        message = `${field.label}不能小于 ${constraint.min}`
      } else if (constraint.max !== undefined && number > constraint.max) {