fix: restructuring safe websocket client

internal/hub.go

@@ -11,7 +11,7 @@ import (
 const (
 	writeWait  = 10 * time.Second
 	pongWait   = 60 * time.Second
-	pingPeriod = 55 * time.Second
+	pingPeriod = 25 * time.Second
 )
 
 type Client struct {
@@ -95,7 +95,7 @@ func WritePump(c *Client, h *Hub) {
 			}
 		case <-pingTicker.C:
 			c.Conn.SetWriteDeadline(time.Now().Add(writeWait))
-			if err := c.Conn.WriteMessage(websocket.PingMessage, nil); err != nil {
+			if err := c.Conn.WriteMessage(websocket.PingMessage, []byte{}); err != nil {
 				return
 			}
 		}

v1/client/client.go

@@ -6,6 +6,7 @@ import (
 	"log"
 	"net/url"
 	"strings"
+	"sync"
 	"time"
 
 	custom_rwmutex "git.neurocipta.com/rogerferdinan/custom-rwmutex"
@@ -17,6 +18,52 @@ const (
 	pingPeriod = 10 * time.Second
 )
 
+type SafeMap[K comparable, V any] struct {
+	m sync.Map
+}
+
+func NewSafeMap[K comparable, V any]() *SafeMap[K, V] {
+	return &SafeMap[K, V]{
+		m: sync.Map{},
+	}
+}
+
+func (sm *SafeMap[K, V]) Store(key K, value V) {
+	sm.m.Store(key, value)
+}
+
+func (sm *SafeMap[K, V]) Load(key K) (value V, ok bool) {
+	val, loaded := sm.m.Load(key)
+	if !loaded {
+		return *new(V), false
+	}
+	return val.(V), true
+}
+
+func (sm *SafeMap[K, V]) Delete(key K) {
+	sm.m.Delete(key)
+}
+
+func (sm *SafeMap[K, V]) Range(f func(K, V) bool) {
+	sm.m.Range(func(key, value any) bool {
+		k, ok1 := key.(K)
+		v, ok2 := value.(V)
+		if !ok1 || !ok2 {
+			return true
+		}
+		return f(k, v)
+	})
+}
+
+func (sm *SafeMap[K, V]) Len() int {
+	count := 0
+	sm.Range(func(_ K, _ V) bool {
+		count++
+		return true
+	})
+	return count
+}
+
 type SafeWebsocketClientBuilder struct {
 	baseHost *string `nil_checker:"required"`
 	basePort *uint16 `nil_checker:"required"`
@@ -59,8 +106,6 @@ func (b *SafeWebsocketClientBuilder) Build() (*SafeWebsocketClient, error) {
 		return nil, err
 	}
 
-	ctx, cancel := context.WithCancel(context.Background())
-
 	var useTLS bool
 	if b.useTLS != nil {
 		useTLS = *b.useTLS
@@ -72,16 +117,16 @@ func (b *SafeWebsocketClientBuilder) Build() (*SafeWebsocketClient, error) {
 		useTLS:      useTLS,
 		path:        b.path,
 		rawQuery:    b.rawQuery,
-		ctx:         ctx,
-		cancel:      cancel,
 		dataChannel: make(chan []byte, 1),
 		mu:          custom_rwmutex.NewCustomRwMutex(),
 		reconnectCh: make(chan struct{}, 1),
 		isConnected: false,
+		doneMap:     NewSafeMap[string, chan struct{}](),
 	}
 
+	go wsClient.reconnectHandler()
+
 	if err := wsClient.connect(); err != nil {
-		cancel()
 		return nil, fmt.Errorf("failed to establish initial connection: %v", err)
 	}
 
@@ -101,6 +146,8 @@ type SafeWebsocketClient struct {
 	cancel      context.CancelFunc
 	reconnectCh chan struct{}
 	isConnected bool
+
+	doneMap *SafeMap[string, chan struct{}]
 }
 
 func (wsClient *SafeWebsocketClient) connect() error {
@@ -129,75 +176,90 @@ func (wsClient *SafeWebsocketClient) connect() error {
 	}
 
 	conn.SetPingHandler(func(pingData string) error {
-		if err := conn.WriteMessage(websocket.PongMessage, []byte(pingData)); err != nil {
-			if err == websocket.ErrCloseSent {
-				return nil
+		return wsClient.mu.WriteHandler(func() error {
+			if err := conn.WriteMessage(websocket.PongMessage, []byte(pingData)); err != nil {
+				if err == websocket.ErrCloseSent {
+					return nil
+				}
+				if netErr, ok := err.(interface{ Timeout() bool }); ok && netErr.Timeout() {
+					return nil
+				}
+				return err
 			}
-			if netErr, ok := err.(interface{ Timeout() bool }); ok && netErr.Timeout() {
-				return nil
-			}
-			return err
-		}
-		return nil
+			return nil
+		})
 	})
 
 	wsClient.mu.WriteHandler(func() error {
+		if wsClient.conn != nil {
+			wsClient.conn.Close()
+		}
+
+		ctx, cancel := context.WithCancel(context.Background())
+		wsClient.ctx = ctx
+		wsClient.cancel = cancel
 		wsClient.conn = conn
 		wsClient.isConnected = true
+
+		go wsClient.startPingTicker(ctx)
+		go wsClient.startReceiveHandler(ctx)
+
 		return nil
 	})
-
-	go wsClient.startPingTicker()
-	go wsClient.startReceiveHandler()
-	go wsClient.reconnectHandler()
-
 	return nil
 }
 
-func (wsClient *SafeWebsocketClient) startPingTicker() {
+func (wsClient *SafeWebsocketClient) startPingTicker(ctx context.Context) {
 	ticker := time.NewTicker(pingPeriod)
 	defer ticker.Stop()
+
 	for {
 		select {
 		case <-ticker.C:
 			wsClient.mu.WriteHandler(func() error {
+				if wsClient.conn == nil {
+					return fmt.Errorf("connecrtion closed")
+				}
 				if err := wsClient.conn.WriteMessage(websocket.PingMessage, []byte{}); err != nil {
 					log.Printf("Ping failed: %v. Will attempt reconnect.", err)
 					wsClient.triggerReconnect()
 				}
 				return nil
 			})
-		case <-wsClient.ctx.Done():
-			log.Println("Ping ticker stopped")
+		case <-ctx.Done():
+			log.Println("Ping ticker stopped due to context cancellation")
 			return
 		}
 	}
 }
 
-func (wsClient *SafeWebsocketClient) startReceiveHandler() {
+func (wsClient *SafeWebsocketClient) startReceiveHandler(ctx context.Context) {
 	for {
 		select {
-		case <-wsClient.reconnectCh:
-		case <-wsClient.ctx.Done():
+		case <-ctx.Done():
 			log.Println("Reconnect handler stopped")
 			return
 		default:
-			if err := wsClient.mu.ReadHandler(func() error {
+			wsClient.mu.ReadHandler(func() error {
 				conn := wsClient.conn
 
 				if conn == nil {
-					return fmt.Errorf("no active connection, waiting for reconnect")
+					wsClient.triggerReconnect()
+					return fmt.Errorf("connection closed")
 				}
 				_, message, err := conn.ReadMessage()
 				if err != nil {
-					return err
+					wsClient.triggerReconnect()
+					return fmt.Errorf("failed to read message: %v", err)
+				}
+				select {
+				case wsClient.dataChannel <- message:
+				default:
+					log.Println("Data channel full, dropping message")
 				}
-				wsClient.dataChannel <- message
 				return nil
-			}); err != nil {
-				wsClient.triggerReconnect()
-				return
-			}
+			})
+
 		}
 	}
 }
@@ -210,8 +272,42 @@ func (wsClient *SafeWebsocketClient) triggerReconnect() {
 }
 
 func (wsClient *SafeWebsocketClient) reconnectHandler() {
-	for range wsClient.reconnectCh {
-		wsClient.connect()
+	backoff := 1 * time.Second
+	maxBackoff := 30 * time.Second
+	for {
+		select {
+		case <-wsClient.reconnectCh:
+			log.Println("Reconnect triggered")
+			wsClient.mu.WriteHandler(func() error {
+				if wsClient.cancel != nil {
+					wsClient.cancel()
+				}
+				wsClient.isConnected = false
+				return nil
+			})
+
+			time.Sleep(100 * time.Millisecond)
+
+			for {
+				log.Println("Attempting reconnect in %v...", backoff)
+				select {
+				case <-time.After(backoff):
+					if err := wsClient.connect(); err != nil {
+						log.Println("Reconnect failed: %v", err)
+						if backoff < maxBackoff {
+							backoff *= 2
+						}
+						continue
+					}
+					log.Println("Reconnected successfully")
+					backoff = 1 * time.Second
+					break
+				}
+			}
+		case <-wsClient.ctx.Done():
+			log.Println("Reconnect handler stopped due to client shutdown")
+			return
+		}
 	}
 }