fix: fixing deadlock caused by read mutex

v1/client/client.go

@@ -6,6 +6,7 @@ import (
 	"log"
 	"net/url"
 	"strings"
+	"sync"
 	"time"
 
 	custom_rwmutex "git.neurocipta.com/rogerferdinan/custom-rwmutex"
@@ -17,12 +18,59 @@ 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"`
-	path     *string
-	rawQuery *string
-	useTLS   *bool
+	baseHost    *string `nil_checker:"required"`
+	basePort    *uint16 `nil_checker:"required"`
+	path        *string
+	rawQuery    *string
+	useTLS      *bool
+	channelSize *int64
 }
 
 func NewSafeWebsocketClientBuilder() *SafeWebsocketClientBuilder {
@@ -54,7 +102,12 @@ func (b *SafeWebsocketClientBuilder) RawQuery(rawQuery string) *SafeWebsocketCli
 	return b
 }
 
-func (b *SafeWebsocketClientBuilder) Build() (*SafeWebsocketClient, error) {
+func (b *SafeWebsocketClientBuilder) ChannelSize(channelSize int64) *SafeWebsocketClientBuilder {
+	b.channelSize = &channelSize
+	return b
+}
+
+func (b *SafeWebsocketClientBuilder) Build(ctx context.Context) (*SafeWebsocketClient, error) {
 	if err := internal.NilChecker(b); err != nil {
 		return nil, err
 	}
@@ -64,18 +117,27 @@ func (b *SafeWebsocketClientBuilder) Build() (*SafeWebsocketClient, error) {
 		useTLS = *b.useTLS
 	}
 
+	if b.channelSize == nil {
+		channelSize := int64(1)
+		b.channelSize = &channelSize
+	}
+
 	wsClient := SafeWebsocketClient{
 		baseHost:    *b.baseHost,
 		basePort:    *b.basePort,
 		useTLS:      useTLS,
 		path:        b.path,
 		rawQuery:    b.rawQuery,
-		dataChannel: make(chan []byte, 1),
+		dataChannel: make(chan []byte, *b.channelSize),
 		mu:          custom_rwmutex.NewCustomRwMutex(),
+		ctx:         ctx,
 		reconnectCh: make(chan struct{}, 1),
 		isConnected: false,
+		doneMap:     NewSafeMap[string, chan struct{}](),
 	}
 
+	go wsClient.reconnectHandler()
+
 	if err := wsClient.connect(); err != nil {
 		return nil, fmt.Errorf("failed to establish initial connection: %v", err)
 	}
@@ -84,18 +146,20 @@ func (b *SafeWebsocketClientBuilder) Build() (*SafeWebsocketClient, error) {
 }
 
 type SafeWebsocketClient struct {
-	baseHost    string
-	basePort    uint16
-	useTLS      bool
-	path        *string
-	rawQuery    *string
-	dataChannel chan []byte
-	mu          *custom_rwmutex.CustomRwMutex
-	conn        *websocket.Conn
-	ctx         context.Context
-	cancel      context.CancelFunc
-	reconnectCh chan struct{}
-	isConnected bool
+	baseHost       string
+	basePort       uint16
+	useTLS         bool
+	path           *string
+	rawQuery       *string
+	dataChannel    chan []byte
+	mu             *custom_rwmutex.CustomRwMutex
+	conn           *websocket.Conn
+	cancelFuncs    []context.CancelFunc
+	ctx            context.Context
+	reconnectCh    chan struct{}
+	reconnectChans []chan struct{}
+	isConnected    bool
+	doneMap        *SafeMap[string, chan struct{}]
 }
 
 func (wsClient *SafeWebsocketClient) connect() error {
@@ -124,84 +188,93 @@ func (wsClient *SafeWebsocketClient) connect() error {
 	}
 
 	conn.SetPingHandler(func(pingData string) error {
-		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 err := conn.WriteMessage(websocket.PongMessage, []byte(pingData)); err != nil {
+			if err == websocket.ErrCloseSent {
+				return nil
 			}
-			return nil
-		})
-	})
-
-	wsClient.mu.WriteHandler(func() error {
-		ctx, cancel := context.WithCancel(context.Background())
-		wsClient.ctx = ctx
-		wsClient.cancel = cancel
-		wsClient.conn = conn
-		wsClient.isConnected = true
-
+			if netErr, ok := err.(interface{ Timeout() bool }); ok && netErr.Timeout() {
+				return nil
+			}
+			return err
+		}
 		return nil
 	})
 
-	go wsClient.startPingTicker()
-	go wsClient.startReceiveHandler()
-	go wsClient.reconnectHandler()
+	wsClient.mu.WriteHandler(func() error {
+		if wsClient.conn != nil {
+			wsClient.conn.Close()
+		}
 
+		wsClient.conn = conn
+
+		wsClient.isConnected = true
+
+		pingCtx, pingCancel := context.WithCancel(context.Background())
+		go wsClient.startPingTicker(pingCtx)
+		wsClient.cancelFuncs = append(wsClient.cancelFuncs, pingCancel)
+
+		receiverCtx, receiverCancel := context.WithCancel(context.Background())
+		go wsClient.startReceiveHandler(receiverCtx)
+		wsClient.cancelFuncs = append(wsClient.cancelFuncs, receiverCancel)
+
+		return nil
+	})
 	return nil
 }
 
-func (wsClient *SafeWebsocketClient) startPingTicker() {
+func (wsClient *SafeWebsocketClient) startPingTicker(ctx context.Context) {
 	ticker := time.NewTicker(pingPeriod)
 	defer ticker.Stop()
+
 	for {
 		select {
+		case <-ctx.Done():
+			log.Println("ping ticker stopped")
+			return
 		case <-ticker.C:
 			wsClient.mu.WriteHandler(func() error {
+				if wsClient.conn == nil {
+					return fmt.Errorf("connection 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")
-			return
 		}
 	}
 }
 
-func (wsClient *SafeWebsocketClient) startReceiveHandler() {
+func (wsClient *SafeWebsocketClient) startReceiveHandler(ctx context.Context) {
 	for {
 		select {
-		case <-wsClient.ctx.Done():
-			log.Println("Reconnect handler stopped")
+		case <-ctx.Done():
+			log.Println("receive handler stopped")
 			return
 		default:
-			// if err := wsClient.mu.ReadHandler(func() error {
 			conn := wsClient.conn
 
 			if conn == nil {
 				wsClient.triggerReconnect()
 				return
-				// return fmt.Errorf("no active connection, waiting for reconnect")
+				// return fmt.Errorf("connection closed")
 			}
 			_, message, err := conn.ReadMessage()
 			if err != nil {
 				wsClient.triggerReconnect()
+				// return fmt.Errorf("failed to read message: %v", err)
 				return
-				// return err
 			}
-			wsClient.dataChannel <- message
-			// return nil
-			// }); err != nil {
-			// 	wsClient.triggerReconnect()
-			// 	return
-			// }
+			select {
+			case wsClient.dataChannel <- message:
+			case <-ctx.Done():
+				log.Println("Reconnect handler stopped")
+				return
+			default:
+				log.Println("Data channel full, dropping message")
+			}
+			return
 		}
 	}
 }
@@ -214,17 +287,93 @@ func (wsClient *SafeWebsocketClient) triggerReconnect() {
 }
 
 func (wsClient *SafeWebsocketClient) reconnectHandler() {
+	backoff := 1 * time.Second
+	maxBackoff := 30 * time.Second
 	for {
 		select {
 		case <-wsClient.reconnectCh:
-			wsClient.cancel()
-			wsClient.connect()
+			log.Println("Reconnect triggered")
+			if wsClient.cancelFuncs != nil {
+				for _, cancel := range wsClient.cancelFuncs {
+					cancel()
+				}
+			}
+			wsClient.isConnected = false
+			time.Sleep(100 * time.Millisecond)
+
+			isInnerLoop := true
+			for isInnerLoop {
+				log.Printf("Attempting reconnect in %v...", backoff)
+				select {
+				case <-time.After(backoff):
+					if err := wsClient.connect(); err != nil {
+						log.Printf("Reconnect failed: %v", err)
+						if backoff < maxBackoff {
+							backoff *= 2
+						}
+						continue
+					}
+					log.Println("Reconnected successfully")
+					backoff = 1 * time.Second
+					isInnerLoop = false
+					continue
+				case <-wsClient.ctx.Done():
+					return
+				}
+			}
+			if wsClient.reconnectChans != nil {
+				for _, reconnectCh := range wsClient.reconnectChans {
+					reconnectCh <- struct{}{}
+				}
+			}
 		case <-wsClient.ctx.Done():
+			log.Println("reconnect handler stopped due to client shutdown")
+			wsClient.Close()
 			return
 		}
 	}
 }
 
+func (wsClient *SafeWebsocketClient) ReconnectChannel() <-chan struct{} {
+	reconnectCh := make(chan struct{})
+	wsClient.mu.WriteHandler(func() error {
+		wsClient.reconnectChans = append(wsClient.reconnectChans, reconnectCh)
+		return nil
+	})
+
+	return reconnectCh
+}
+
 func (wsClient *SafeWebsocketClient) DataChannel() <-chan []byte {
 	return wsClient.dataChannel
 }
+
+func (wsClient *SafeWebsocketClient) WriteJSON(message any) error {
+	return wsClient.mu.WriteHandler(func() error {
+		return wsClient.conn.WriteJSON(message)
+	})
+}
+
+func (wsClient *SafeWebsocketClient) Close() error {
+	wsClient.mu.WriteHandler(func() error {
+		if wsClient.cancelFuncs != nil {
+			for _, cancel := range wsClient.cancelFuncs {
+				cancel()
+			}
+		}
+
+		if wsClient.reconnectChans != nil {
+			for _, reconnectChan := range wsClient.reconnectChans {
+				close(reconnectChan)
+			}
+		}
+		if wsClient.conn != nil {
+			wsClient.conn.Close()
+		}
+		wsClient.isConnected = false
+		return nil
+	})
+	close(wsClient.dataChannel)
+
+	return nil
+}

v1/examples/client/main.go

@@ -1,22 +1,45 @@
 package main
 
 import (
+	"context"
 	"fmt"
 	"log"
+	"os"
+	"os/signal"
+	"syscall"
 
 	"git.neurocipta.com/rogerferdinan/safe-web-socket/v1/client"
 )
 
 func main() {
+	sigChan := make(chan os.Signal, 1)
+	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	go func() {
+		<-sigChan
+		fmt.Println("\nReceived interrupt signal. Shutting down gracefully...")
+		cancel()
+	}()
+
 	wsClient, err := client.NewSafeWebsocketClientBuilder().
 		BaseHost("localhost").
 		BasePort(8080).
 		Path("/ws/test/data_1").
 		UseTLS(false).
-		Build()
+		Build(ctx)
 	if err != nil {
 		log.Fatal(err)
 	}
+
+	go func() {
+		for range wsClient.ReconnectChannel() {
+			fmt.Println("Reconnection Success")
+		}
+	}()
+
 	dataChannel := wsClient.DataChannel()
 
 	for data := range dataChannel {