Automatically reconnect if keepalive keeps failing

keepalive.go

@@ -23,10 +23,13 @@ var (
 	KeepAliveIntervalMin = 20 * time.Second
 	// KeepAliveIntervalMax specifies the maximum interval for websocket keepalive pings.
 	KeepAliveIntervalMax = 30 * time.Second
+
+	// KeepAliveMaxFailTime specifies the maximum time to wait before forcing a reconnect if keepalives fail repeatedly.
+	KeepAliveMaxFailTime = 3 * time.Minute
 )
 
 func (cli *Client) keepAliveLoop(ctx context.Context) {
-	var lastSuccess time.Time
+	lastSuccess := time.Now()
 	var errorCount int
 	for {
 		interval := rand.Int63n(KeepAliveIntervalMax.Milliseconds()-KeepAliveIntervalMin.Milliseconds()) + KeepAliveIntervalMin.Milliseconds()
@@ -41,6 +44,11 @@ func (cli *Client) keepAliveLoop(ctx context.Context) {
 					ErrorCount:  errorCount,
 					LastSuccess: lastSuccess,
 				})
+				if cli.EnableAutoReconnect && time.Since(lastSuccess) > KeepAliveMaxFailTime {
+					cli.Log.Debugf("Forcing reconnect due to keepalive failure")
+					cli.Disconnect()
+					go cli.autoReconnect()
+				}
 			} else {
 				if errorCount > 0 {
 					errorCount = 0

mdtest/main.go

@@ -843,16 +843,6 @@ func handler(rawEvt interface{}) {
 		log.Debugf("App state event: %+v / %+v", evt.Index, evt.SyncActionValue)
 	case *events.KeepAliveTimeout:
 		log.Debugf("Keepalive timeout event: %+v", evt)
-		if evt.ErrorCount > 3 {
-			log.Debugf("Got >3 keepalive timeouts, forcing reconnect")
-			go func() {
-				cli.Disconnect()
-				err := cli.Connect()
-				if err != nil {
-					log.Errorf("Error force-reconnecting after keepalive timeouts: %v", err)
-				}
-			}()
-		}
 	case *events.KeepAliveRestored:
 		log.Debugf("Keepalive restored")
 	}