From 34190f5a04d8554961eba7ea3e687bca200454e1 Mon Sep 17 00:00:00 2001
From: JustMe <mr.toing@outlook.com>
Date: Sun, 18 Oct 2026 00:23:48 +0800
Subject: [PATCH] fix(link): bound the server connect and alternate address
 families on retry [CI]

---
 backend/fleet/src/Fleet.Link/FleetLink.cs | 61 +++++++++++++++++++++++
 1 file changed, 61 insertions(+)

diff --git a/backend/fleet/src/Fleet.Link/FleetLink.cs b/backend/fleet/src/Fleet.Link/FleetLink.cs
index 07755c2..607f93f 100644
--- a/backend/fleet/src/Fleet.Link/FleetLink.cs
+++ b/backend/fleet/src/Fleet.Link/FleetLink.cs
@@ -1,3 +1,6 @@
+using System.Net;
+using System.Net.Http;
+using System.Net.Sockets;
 using System.Net.WebSockets;
 using System.Text.Json;
 using System.Text.Json.Serialization.Metadata;
@@ -93,6 +96,9 @@ public sealed class FleetLink
     private readonly INodeLog _log;
     private volatile RpcSession<LinkContext>? _session;
     private volatile bool _stopped;
+    private static readonly TimeSpan s_connectTimeout = TimeSpan.FromSeconds(20);
+    private static readonly TimeSpan s_tcpTimeout = TimeSpan.FromSeconds(4);
+    private int _failedConnects;
 
     public FleetLink(FleetLinkOptions options, RpcRouter<LinkContext> router, IEnumerable<IStateHandler> handlers, Outbox outbox, TimeProvider time, INodeLog log)
     {
@@ -163,6 +169,61 @@ public sealed class FleetLink
         }
     }
 
+    /// <summary>
+    /// 连上面板：整个握手（TCP、TLS、升级）限时，卡住就放弃重试；每个地址的 TCP 连接也单独限时。
+    /// 失败之后下一次换着先试另一个地址族（IPv6 / IPv4）：有的网络上到某个地址族的 TCP 能连上、之后的 TLS 却被卡住，换一个往往就通。
+    /// </summary>
+    private async Task ConnectAsync(ClientWebSocket socket, Uri uri, CancellationToken ct)
+    {
+        int attempt = Volatile.Read(ref _failedConnects);
+        using var handler = new SocketsHttpHandler { ConnectCallback = (context, token) => ConnectTcpAsync(context, attempt, token) };
+        using var invoker = new HttpMessageInvoker(handler, disposeHandler: false);
+        using CancellationTokenSource timeout = CancellationTokenSource.CreateLinkedTokenSource(ct);
+        timeout.CancelAfter(s_connectTimeout);
+        try
+        {
+            await socket.ConnectAsync(uri, invoker, timeout.Token);
+            Volatile.Write(ref _failedConnects, 0);
+        }
+        catch (OperationCanceledException) when (!ct.IsCancellationRequested)
+        {
+            Interlocked.Increment(ref _failedConnects);
+            throw new TimeoutException("connecting to the server timed out");
+        }
+        catch
+        {
+            Interlocked.Increment(ref _failedConnects);
+            throw;
+        }
+    }
+
+    private static async ValueTask<Stream> ConnectTcpAsync(SocketsHttpConnectionContext context, int attempt, CancellationToken ct)
+    {
+        IPAddress[] addresses = await Dns.GetHostAddressesAsync(context.DnsEndPoint.Host, ct);
+        // 第一次按 DNS 的顺序；之后奇数次先试 IPv6，偶数次先试 IPv4
+        IEnumerable<IPAddress> ordered = attempt == 0 ? addresses
+            : addresses.OrderBy(a => (a.AddressFamily == AddressFamily.InterNetworkV6) == (attempt % 2 == 1) ? 0 : 1);
+        Exception? last = null;
+        foreach (IPAddress address in ordered)
+        {
+            var tcp = new Socket(address.AddressFamily, SocketType.Stream, ProtocolType.Tcp) { NoDelay = true };
+            using CancellationTokenSource limit = CancellationTokenSource.CreateLinkedTokenSource(ct);
+            limit.CancelAfter(s_tcpTimeout);
+            try
+            {
+                await tcp.ConnectAsync(new IPEndPoint(address, context.DnsEndPoint.Port), limit.Token);
+                return new NetworkStream(tcp, ownsSocket: true);
+            }
+            catch (Exception ex) when (ex is SocketException or OperationCanceledException && !ct.IsCancellationRequested)
+            {
+                tcp.Dispose();
+                last = ex;
+            }
+        }
+
+        throw last ?? new SocketException((int)SocketError.HostNotFound);
+    }
+
     /// <summary>当前这条链路是不是应用层加密的（服务器证明了身份）。</summary>
     public bool IsSecure => _secure && IsConnected;