Infinite reconnecting to a channel on Android/iOS platforms

[Fugasss]

Unity version: 6000.3.8f1
PhoenixSharp version: any version from January 2026 (tried 1.0.1, 1.2.1, 1.3.1 installed using NuGet)
.net version: 2.1
Platforms: Android/iOS with IL2CPP enabled

I planned to use async methods but after updating of this plugin from version released in 2023, I got the problem with websocket behavior on Android/iOS devices while in Editor all works as intended.

In total I have 2 different channels, but both of them act similar: phx_join, phx_reply, phx_close with smaller Ref value. Then channel is reconnecting infinitely with the same problem.

Logs for 1 of channels:

Pushing Message { Topic = room:1, Event = phx_join, Ref = 8, Payload = Phoenix.JsonBox, JoinRef = 8 }
Received Message { Topic = room:1, Event = phx_reply, Ref = 8, Payload = Phoenix.JsonBox, JoinRef = 8 }
Received Message { Topic = room:1, Event = phx_close, Ref = 6, Payload = Phoenix.JsonBox, JoinRef = 6 }
dropping outdated message for topic 'room:1' (joinRef 6 does not match joinRef 8)

Pushing Message { Topic = room:1, Event = phx_join, Ref = 12, Payload = Phoenix.JsonBox, JoinRef = 12 }
Received Message { Topic = room:1, Event = phx_reply, Ref = 12, Payload = Phoenix.JsonBox, JoinRef = 12 }
Received Message { Topic = room:1, Event = phx_close, Ref = 10, Payload = Phoenix.JsonBox, JoinRef = 10 }
dropping outdated message for topic 'room:1' (joinRef 10 does not match joinRef 12)

and so on

[Mazyod]

Thanks for the detailed report! It's extremely strange that it only happens on a single platform. Can you please share with me what are you using for the Websocket transport in C#? I personally use BestHTTP library, and tested everything on physical iOS hardware, everything working pretty well.

Sharing with me the specifics of the web socket implementation in particular will allow me to hopefully reproduce easily.

In all cases, will take a look and see.

[Fugasss]

I tried WebSocketSharp and System.Net.WebSockets.ClientWebSocket, but both of them gave the same result. Downgrading to the old version of plugin fixed the issue.
Also, I'm using UniTask for delayed execution.

using System;
using System.Threading;
using Cysharp.Threading.Tasks;
using Phoenix;

namespace Core.Network.PhoenixWebsocket
{
    public sealed class UniTaskExecution : IDelayedExecution
    {
        private bool _cancelled;
        private CancellationTokenSource _cts;

        internal UniTaskExecution(CancellationTokenSource cts)
        {
            _cts = cts;
        }

        public void Cancel()
        {
            if (_cancelled) return;

            _cancelled = true;
            _cts.Cancel();
            _cts.Dispose();
        }
    }
    
    public sealed class UniTaskDelayedExecutor : IDelayedExecutor
    {
        public IDelayedExecution Execute(Action action, TimeSpan delay)
        {
            var cts = new CancellationTokenSource();
            var execution = new UniTaskExecution(cts);

            DelayAndInvokeAsync(action, delay, cts.Token).Forget();

            return execution;
        }

        private async UniTaskVoid DelayAndInvokeAsync(Action action, TimeSpan delay, CancellationToken token)
        {
            try
            {
                await UniTask.Delay(delay, cancellationToken: token);
                if (!token.IsCancellationRequested)
                {
                    action?.Invoke();
                }
            }
            catch (OperationCanceledException)
            {
            }
        }
    }
}

[Mazyod]

Hey @Fugasss, thanks so much for reporting this — it ended up leading to several meaningful fixes!

After digging deep into the codebase and comparing against the PhoenixJS reference implementation, we found two concrete bugs in the reconnection code that were introduced in the recent refactor:

1. Leaked heartbeat timer executions — ResetHeartbeat() and the heartbeat response handler were scheduling SendHeartbeat without storing the reference, so they could never be cancelled. After reconnection, orphaned timer callbacks would fire on the new connection, causing spurious closes.

2. Heartbeat timeout not triggering reconnection properly — HeartbeatTimeout() was calling AbnormalClose() which relied on OnConnClose, but the code != 1000 check there prevented reconnection. Rewrote it to match PhoenixJS's approach: trigger channel errors, then teardown with an explicit reconnect callback.

On top of that, we also shipped a NativeWebSocket sample adapter as a modern replacement for WebSocketSharp (which has TLS issues and is no longer maintained). While testing it on a real Android device (Pixel 8a, IL2CPP) via Firebase Test Lab, we discovered that NativeWebSocket queues received messages on all platforms — without calling DispatchMessageQueue() from Update(), channel replies never arrive and you get exactly the infinite rejoin loop you described. The sample adapter now includes a Tick() method with clear documentation on why it's needed.

All of this is included in v1.4.0, which is now published to both NuGet and OpenUPM.

Would love to hear if this resolves the issue on your end! If you're still seeing the loop after updating, please reopen and we'll dig further. Any additional details about your WebSocket adapter setup would be super helpful in that case.

Thanks again for taking the time to report this 🙏