fix(link): ride out short path blackouts: 30s handshake, 180s idle limit, 90s keepalive, faster probe reconnect [CI]

backend/fleet/src/Fleet.Link/Peer.cs

@@ -29,7 +29,8 @@ public sealed class PeerListenerOptions
 
     public int MaxSessions { get; init; } = 8;
 
-    public TimeSpan HandshakeTimeout { get; init; } = TimeSpan.FromSeconds(10);
+    /// <summary>敲门、TLS、挑战应答整个握手的上限。线路差的时候（RTT 好几秒、还丢包）握手本来就要很久，10 秒做不完就会一直连不上；敲门已经挡住了不认识的连接，放宽不会放大被占连接的风险。</summary>
+    public TimeSpan HandshakeTimeout { get; init; } = PeerTimings.Handshake;
 }
 
 /// <summary>
@@ -209,7 +210,8 @@ public sealed class PeerListener
 }
 
 /// <summary>发起点对点会话（Probe 一侧）：按 SPKI 指纹固定对端证书，用身份私钥签名，派生同样的数据面凭据。</summary>
-/// <summary>点对点会话的套接字：开启 TCP keepalive（空闲 15 秒后每 5 秒探测，连续 3 次无响应就断开，约 30 秒），主机丢失或路径被切断时连接不会一直挂着。</summary>
+/// <summary>点对点会话的套接字：开启 TCP keepalive（空闲 30 秒后每 10 秒探测，连续 6 次无响应才断开，约 90 秒），主机真的没了的时候连接不会一直挂着；
+/// 但线路短暂中断（几十秒的黑洞）不能把会话杀掉——线路中断本来就是要测的东西，会话得扛过去，恢复之后接着测。</summary>
 internal static class PeerSockets
 {
     public static void EnableKeepAlive(Socket socket)
@@ -217,9 +219,9 @@ internal static class PeerSockets
         try
         {
             socket.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.KeepAlive, true);
-            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveTime, 15);
-            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveInterval, 5);
-            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveRetryCount, 3);
+            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveTime, 30);
+            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveInterval, 10);
+            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveRetryCount, 6);
         }
         catch (Exception ex) when (ex is SocketException or ArgumentException or PlatformNotSupportedException)
         {
@@ -228,6 +230,13 @@ internal static class PeerSockets
     }
 }
 
+/// <summary>点对点会话的时间常数，Probe 和 Agent 两侧共用。</summary>
+public static class PeerTimings
+{
+    /// <summary>建立会话（连接、敲门、TLS、挑战应答）的总时限。</summary>
+    public static readonly TimeSpan Handshake = TimeSpan.FromSeconds(30);
+}
+
 public static class PeerConnector
 {
     public static async Task<(PeerSessionInfo Info, TlsFrameTransport Transport)> ConnectAsync(
@@ -240,7 +249,7 @@ public static class PeerConnector
         PeerSockets.EnableKeepAlive(client.Client);
         try
         {
-            using var timeout = new CancellationTokenSource(TimeSpan.FromSeconds(10), time);
+            using var timeout = new CancellationTokenSource(PeerTimings.Handshake, time);
             using CancellationTokenSource linked = CancellationTokenSource.CreateLinkedTokenSource(ct, timeout.Token);
             await client.ConnectAsync(host, port, linked.Token);
             Stream stream = client.GetStream();