v1.5.1 增强程序稳定性

EdgeDNS/internal/nodes/dns_node.go

@@ -34,6 +34,7 @@ import (
 	"regexp"
 	"runtime"
 	"runtime/debug"
+	"sync/atomic"
 	"syscall"
 	"time"
 )
@@ -46,7 +47,18 @@ var sharedDomainManager *DomainManager
 var sharedRecordManager *RecordManager
 var sharedRouteManager *RouteManager
 var sharedKeyManager *KeyManager
-var sharedNodeConfig = &dnsconfigs.NSNodeConfig{}
+
+var sharedNodeConfig atomic.Pointer[dnsconfigs.NSNodeConfig]
+
+func init() {
+	// 初始化默认空配置
+	sharedNodeConfig.Store(&dnsconfigs.NSNodeConfig{})
+}
+
+// dnsNodeConfig 返回当前 DNS 节点配置（并发安全）
+func dnsNodeConfig() *dnsconfigs.NSNodeConfig {
+	return sharedNodeConfig.Load()
+}
 
 func NewDNSNode() *DNSNode {
 	return &DNSNode{
@@ -105,13 +117,19 @@ func (this *DNSNode) Start() {
 	events.Notify(events.EventStart)
 
 	// 监控状态
-	go NewNodeStatusExecutor().Listen()
+	goman.New(func() {
+		NewNodeStatusExecutor().Listen()
+	})
 
 	// 连接API
-	go NewAPIStream().Start()
+	goman.New(func() {
+		NewAPIStream().Start()
+	})
 
 	// 启动
-	go this.start()
+	goman.New(func() {
+		this.start()
+	})
 
 	// Hold住进程
 	logs.Println("[DNS_NODE]started")
@@ -224,7 +242,7 @@ func (this *DNSNode) listenSock() error {
 	}
 
 	// 启动监听
-	go func() {
+	goman.New(func() {
 		this.sock.OnCommand(func(cmd *gosock.Command) {
 			switch cmd.Code {
 			case "pid":
@@ -262,7 +280,7 @@ func (this *DNSNode) listenSock() error {
 		if err != nil {
 			logs.Println("NODE", err.Error())
 		}
-	}()
+	})
 
 	events.On(events.EventQuit, func() {
 		logs.Println("[DNS_NODE]", "quit unix sock")
@@ -317,9 +335,9 @@ func (this *DNSNode) start() {
 		return
 	}
 
-	sharedNodeConfig = config
+	sharedNodeConfig.Store(config)
 
-	configs.SharedNodeConfig = config
+	configs.SharedNodeConfig.Store(config)
 	events.Notify(events.EventReload)
 
 	sharedNodeConfigManager.reload(config)
@@ -343,29 +361,47 @@ func (this *DNSNode) start() {
 		return
 	}
 
-	go sharedNodeConfigManager.Start()
+	goman.New(func() {
+		sharedNodeConfigManager.Start()
+	})
 
 	sharedDomainManager = NewDomainManager(db, config.ClusterId)
-	go sharedDomainManager.Start()
+	goman.New(func() {
+		sharedDomainManager.Start()
+	})
 
 	sharedRecordManager = NewRecordManager(db)
-	go sharedRecordManager.Start()
+	goman.New(func() {
+		sharedRecordManager.Start()
+	})
 
 	sharedRouteManager = NewRouteManager(db)
-	go sharedRouteManager.Start()
+	goman.New(func() {
+		sharedRouteManager.Start()
+	})
 
 	sharedKeyManager = NewKeyManager(db)
-	go sharedKeyManager.Start()
+	goman.New(func() {
+		sharedKeyManager.Start()
+	})
 
 	agents.SharedManager = agents.NewManager(db)
-	go agents.SharedManager.Start()
+	goman.New(func() {
+		agents.SharedManager.Start()
+	})
 
-	// 发送通知，这里发送通知需要在DomainManager、RecordeManager等加载完成之后
-	time.Sleep(1 * time.Second)
+	// 等待所有 Manager 初始加载完成后再发送通知
+	<-sharedDomainManager.readyCh
+	<-sharedRecordManager.readyCh
+	<-sharedRouteManager.readyCh
+	<-sharedKeyManager.readyCh
+	<-agents.SharedManager.ReadyCh
 	events.Notify(events.EventLoaded)
 
 	// 启动循环
-	go this.loop()
+	goman.New(func() {
+		this.loop()
+	})
 }
 
 // 更新配置Loop
@@ -439,8 +475,9 @@ func (this *DNSNode) updateDDoS(rpcClient *rpc.RPCClient) error {
 		return err
 	}
 	if len(resp.DdosProtectionJSON) == 0 {
-		if sharedNodeConfig != nil {
-			sharedNodeConfig.DDoSProtection = nil
+		var cfg = dnsNodeConfig()
+		if cfg != nil {
+			cfg.DDoSProtection = nil
 		}
 	} else {
 		var ddosProtectionConfig = &ddosconfigs.ProtectionConfig{}
@@ -449,8 +486,9 @@ func (this *DNSNode) updateDDoS(rpcClient *rpc.RPCClient) error {
 			return fmt.Errorf("decode DDoS protection config failed: %w", err)
 		}
 
-		if sharedNodeConfig != nil {
-			sharedNodeConfig.DDoSProtection = ddosProtectionConfig
+		var cfg = dnsNodeConfig()
+		if cfg != nil {
+			cfg.DDoSProtection = ddosProtectionConfig
 		}
 
 		err = firewalls.SharedDDoSProtectionManager.Apply(ddosProtectionConfig)