fix(nodes): locate agents behind a proxy and retry missing countries [CI]

backend/fleet/src/Fleet.Server.Abstractions/ServerContracts.cs

@@ -6,6 +6,17 @@ namespace Fleet.Server;
 
 public sealed record EnrollmentContext(string? RemoteIp, JsonElement? ProductMeta);
 
+/// <summary>
+/// 节点的真实来源地址：前面有反向代理时，套接字上只有代理的地址，产品在这里决定信不信转发头。
+/// 没有注册就用套接字地址。
+/// </summary>
+public interface IClientAddressSource
+{
+    /// <param name="socket">套接字上看到的对端地址。</param>
+    /// <param name="header">按名字取请求头的所有值。</param>
+    string? Resolve(System.Net.IPAddress? socket, Func<string, IEnumerable<string>> header);
+}
+
 /// <summary>
 /// 服务器的身份密钥：操作员用 passkey 把它的公钥签进操作员集合。有它的时候，认证挑战会带上服务器的签名，节点据此确认对面是真的服务器，
 /// 并在这条链路上用应用层加密。没有注册就是旧的行为（没有服务器身份）。
@@ -38,7 +49,8 @@ public enum NodeEventKind
     Heartbeat,
 }
 
- public sealed record NodeEvent(NodeEventKind Kind, NodeRecord Node, long AtMs, HelloParams? Hello = null, JsonElement? Extra = null, string? Reason = null);
+ /// <param name="RemoteIp">这条连接的来源地址（经过 <see cref="IClientAddressSource"/>）；没有就是 null。</param>
+public sealed record NodeEvent(NodeEventKind Kind, NodeRecord Node, long AtMs, HelloParams? Hello = null, JsonElement? Extra = null, string? Reason = null, string? RemoteIp = null);
 
 public interface INodeEventSink
 {

backend/fleet/src/Fleet.Server.Enrollment/FleetEnrollmentEndpoints.cs

@@ -73,7 +73,7 @@ public static class FleetEnrollmentEndpoints
             return Problem(StatusCodes.Status400BadRequest, FleetErrors.InvalidParams, "request body is not valid");
         }
 
-        Result<EnrollResponse> result = await service.EnrollAsync(request, http.Connection.RemoteIpAddress?.ToString(), http.RequestAborted);
+        Result<EnrollResponse> result = await service.EnrollAsync(request, ClientIp(http), http.RequestAborted);
         if (result.TryGet(out EnrollResponse? response, out FleetError? error))
         {
             return Results.Json(response, EnrollmentJsonContext.Default.EnrollResponse, statusCode: StatusCodes.Status201Created);
@@ -88,6 +88,11 @@ public static class FleetEnrollmentEndpoints
         return Problem(status, error.Code, error.Message);
     }
 
+    private static string? ClientIp(HttpContext http) =>
+        http.RequestServices.GetService<IClientAddressSource>() is { } source
+            ? source.Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? ""))
+            : http.Connection.RemoteIpAddress?.ToString();
+
     private static async Task<EnrollRequest?> ReadBodyAsync(HttpContext http, CancellationToken ct)
     {
         if (http.Request.ContentLength > MaxBodyBytes)

backend/fleet/src/Fleet.Server.Link/FleetLinkHost.cs

@@ -67,9 +67,9 @@ public sealed class FleetLinkHost : IFleetConnections, IStateNotifier
         _connections.TryGetValue(node, out NodeConnection? c) ? c.Session : null;
 
     /// <summary>处理一条 WebSocket 连接，直到连接结束。</summary>
-    public async Task RunConnectionAsync(WebSocket socket, CancellationToken ct)
+    public async Task RunConnectionAsync(WebSocket socket, string? remoteIp, CancellationToken ct)
     {
-        using var connection = new NodeConnection(this, socket);
+        using var connection = new NodeConnection(this, socket, remoteIp);
         await connection.RunAsync(ct);
     }
 

backend/fleet/src/Fleet.Server.Link/FleetServerBuilder.cs

@@ -128,7 +128,10 @@ public static class FleetServerExtensions
             }
 
             using System.Net.WebSockets.WebSocket socket = await http.WebSockets.AcceptWebSocketAsync();
-            await http.RequestServices.GetRequiredService<FleetLinkHost>().RunConnectionAsync(socket, http.RequestAborted);
+            string? remote = http.RequestServices.GetService<IClientAddressSource>() is { } source
+                ? source.Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? ""))
+                : http.Connection.RemoteIpAddress?.ToString();
+            await http.RequestServices.GetRequiredService<FleetLinkHost>().RunConnectionAsync(socket, remote, http.RequestAborted);
         });
         return app;
     }

backend/fleet/src/Fleet.Server.Link/NodeConnection.State.cs

@@ -102,7 +102,7 @@ internal sealed partial class NodeConnection
 
     private async Task FireAsync(NodeEventKind kind, CancellationToken ct, HelloParams? hello = null, JsonElement? extra = null)
     {
-        var nodeEvent = new NodeEvent(kind, _node!, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), hello, extra, kind == NodeEventKind.Offline ? _closeReason : null);
+        var nodeEvent = new NodeEvent(kind, _node!, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), hello, extra, kind == NodeEventKind.Offline ? _closeReason : null, _remoteIp);
         foreach (INodeEventSink sink in _host.Sinks)
         {
             try

backend/fleet/src/Fleet.Server.Link/NodeConnection.cs

@@ -34,10 +34,13 @@ internal sealed partial class NodeConnection : IDisposable
     private bool _helloDone;
     private int _violations;
 
-    public NodeConnection(FleetLinkHost host, WebSocket socket)
+    private readonly string? _remoteIp;
+
+    public NodeConnection(FleetLinkHost host, WebSocket socket, string? remoteIp = null)
     {
         _host = host;
         _socket = socket;
+        _remoteIp = remoteIp;
     }
 
     public NodeId NodeId => _node!.Id;

backend/src/MoShiro.Backend.Api/ApiModule.cs

@@ -20,6 +20,7 @@ public static class ApiModule
         services.AddSingleton<SseLimiter>();
         services.AddSingleton<MoShiro.Backend.Application.IReleaseKey, EmbeddedReleaseKey>();
         services.AddSingleton<ClientAddressResolver>();
+        services.AddSingleton<Fleet.Server.IClientAddressSource>(p => p.GetRequiredService<ClientAddressResolver>());
         services.Configure<JsonOptions>(options =>
         {
             JsonSerializerOptions json = options.SerializerOptions;

backend/src/MoShiro.Backend.Api/Common/ClientAddress.cs

@@ -11,7 +11,7 @@ namespace MoShiro.Backend.Api;
 /// 且从右往左跳过所有可信代理，取第一个不可信的地址（客户端自己伪造的在最左边，不会被采信）。
 /// 没配置就一律用套接字地址：宁可限流和审计里是代理的地址，也不让客户端自己报地址。
 /// </summary>
-public sealed class ClientAddressResolver
+public sealed class ClientAddressResolver : Fleet.Server.IClientAddressSource
 {
     private readonly IPNetwork[] _trusted;
     private readonly string _header;
@@ -43,9 +43,13 @@ public sealed class ClientAddressResolver
         return _trusted.Any(n => n.Contains(normalized));
     }
 
-    public string Resolve(HttpContext http)
+    string? Fleet.Server.IClientAddressSource.Resolve(IPAddress? socket, Func<string, IEnumerable<string>> header) => Resolve(socket, header);
+
+    public string Resolve(HttpContext http) =>
+        Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? ""));
+
+    private string Resolve(IPAddress? remote, Func<string, IEnumerable<string>> headers)
     {
-        IPAddress? remote = http.Connection.RemoteIpAddress;
         if (remote is null)
         {
             return "unknown";
@@ -56,7 +60,7 @@ public sealed class ClientAddressResolver
             return remote.ToString();
         }
 
-        string[] chain = [.. http.Request.Headers[_header].SelectMany(v => (v ?? "").Split(',', StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))];
+        string[] chain = [.. headers(_header).SelectMany(v => (v ?? "").Split(',', StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))];
         for (int i = chain.Length - 1; i >= 0; i--)
         {
             if (!IPAddress.TryParse(chain[i], out IPAddress? hop))

backend/src/MoShiro.Backend.Application/Agents/MoShiroFleetHooks.cs

@@ -115,10 +115,15 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
                 await _events.AddAsync(id, "agent_offline", nodeEvent.AtMs, null, ct);
                 break;
             case NodeEventKind.Hello when nodeEvent.Hello is { } hello:
-                await ApplyHelloAsync(id, hello, ct);
+                await ApplyHelloAsync(id, hello, nodeEvent.RemoteIp, ct);
                 break;
             case NodeEventKind.Heartbeat when nodeEvent.Extra is { } extra:
                 await _nodes.SetVersionAsync(id.Value, NodeKinds.Agent, nodeEvent.Node.NodeVersion, ct);
+                if (_nodes.ShouldRetryGeo(id.Value) && await _agents.GetAsync(id, ct) is { } unlocated)
+                {
+                    _nodes.ObserveAgentInBackground(id.Value, GeoHostOf(unlocated, nodeEvent.RemoteIp), null);
+                }
+
                 try
                 {
                     AgentRuntimeState state = _runtime.Get(id);
@@ -145,6 +150,12 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
         }
     }
 
+    /// <summary>查国家和运营商用哪个地址：对外地址优先，其次 IPv6 地址，都没有（NAT 后面、或者没有上报公网地址）才用这条连接的来源地址。</summary>
+    private static string? GeoHostOf(AgentRecord? agent, string? remoteIp) =>
+        !string.IsNullOrEmpty(agent?.PublicHost) ? agent.PublicHost
+        : !string.IsNullOrEmpty(agent?.Ipv6Host) ? agent.Ipv6Host
+        : remoteIp;
+
     /// <summary>Probe 自己连上来了：hello 里带着它的系统、架构和构建信息；上线下线只推给 SSE，不记事件。</summary>
     private async ValueTask OnProbeEventAsync(NodeId id, NodeEvent nodeEvent, CancellationToken ct)
     {
@@ -213,7 +224,7 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
         }
     }
 
-    private async ValueTask ApplyHelloAsync(NodeId id, HelloParams hello, CancellationToken ct)
+    private async ValueTask ApplyHelloAsync(NodeId id, HelloParams hello, string? remoteIp, CancellationToken ct)
     {
         var kinds = new List<string>();
         foreach (string capability in hello.Capabilities)
@@ -278,7 +289,7 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
 
         // 端口、证书指纹或对外地址变了，Probe 需要新的列表
         AgentRecord? after = await _agents.GetAsync(id, ct);
-        _nodes.ObserveAgentInBackground(id.Value, after?.PublicHost, meta.Platform);
+        _nodes.ObserveAgentInBackground(id.Value, GeoHostOf(after, remoteIp), meta.Platform);
         await _releases.OfferToNodeAsync(id, ct);
         await _schedule.SendToNodeAsync(id, ct);
         if (before is not null && after is not null

backend/src/MoShiro.Backend.Application/Nodes/NodeInfoService.cs

@@ -26,6 +26,7 @@ public sealed class NodeInfoService
     private readonly TimeProvider _time;
     private readonly ConcurrentDictionary<string, NodeMeta> _cache = new();
     private readonly ConcurrentDictionary<string, byte> _inflight = new();
+    private readonly ConcurrentDictionary<string, long> _lastTry = new();
     private Task? _load;
 
     public NodeInfoService(INodeMetaStore store, IGeoLookup geo, TimeProvider time)
@@ -119,6 +120,7 @@ public sealed class NodeInfoService
     public async ValueTask ObserveAddressAsync(string nodeId, string kind, string? host, CancellationToken ct)
     {
         host = host?.Trim();
+        _lastTry[nodeId] = Now();
         if (string.IsNullOrEmpty(host) || !IsPublic(host) || !_inflight.TryAdd(nodeId, 0))
         {
             return;
@@ -162,6 +164,17 @@ public sealed class NodeInfoService
         }
     }
 
+    /// <summary>国家还没查到、并且离上一次尝试已经过了重试间隔：调用方可以再查一次（心跳里调，不会每次都真去查）。</summary>
+    public bool ShouldRetryGeo(string nodeId)
+    {
+        if (_cache.TryGetValue(nodeId, out NodeMeta? meta) && (meta.Info.Country is not null || meta.Info.IspManual))
+        {
+            return false;
+        }
+
+        return !_lastTry.TryGetValue(nodeId, out long at) || Now() - at >= s_retryAfter.TotalMilliseconds;
+    }
+
     /// <summary>Agent 报告某个 Probe 连着它：在后台用 Probe 的来源地址查国家和运营商（系统信息是 Probe 自己连上面板时报的）。出错只吞掉。</summary>
     public void ObserveProbeInBackground(string probeId, string? remote) =>
         _ = Task.Run(async () =>

backend/tests/MoShiro.Backend.Tests/ClientAddressTests.cs

@@ -48,4 +48,12 @@ public sealed class ClientAddressTests
     [Fact]
     public void AMalformedTrustedProxyEntry_FailsAtStartup() =>
         Assert.Throws<InvalidOperationException>(() => new ClientAddressResolver(new BackendOptions { TrustedProxies = "banana" }));
+
+    [Fact]
+    public void TheFleetLinkAsksTheSameResolver_SoANodeBehindAProxyKeepsItsRealAddress()
+    {
+        Fleet.Server.IClientAddressSource source = new ClientAddressResolver(new BackendOptions { TrustedProxies = "127.0.0.1" });
+        Assert.Equal("198.51.100.7", source.Resolve(IPAddress.Parse("127.0.0.1"), _ => ["198.51.100.7"]));
+        Assert.Equal("203.0.113.9", source.Resolve(IPAddress.Parse("203.0.113.9"), _ => ["198.51.100.7"]));
+    }
 }

backend/tests/MoShiro.Backend.Tests/NodeInfoTests.cs

@@ -102,6 +102,24 @@ public sealed partial class NodeInfoTests
         Assert.Null(Assert.Single(await service.ListAsync(CancellationToken.None)).Country);
     }
 
+    [Fact]
+    public async Task Service_AnUnlocatedNode_IsRetriedFromTheHeartbeat_UntilItHasACountry()
+    {
+        var time = new FakeTimeProvider(DateTimeOffset.Parse("2026-10-06T00:00:00Z"));
+        var service = new NodeInfoService(new MemoryStore(), new FakeGeo(null), time);
+
+        Assert.True(service.ShouldRetryGeo("agt_1"));
+        await service.ObserveAddressAsync("agt_1", NodeKinds.Agent, "", CancellationToken.None);
+        Assert.False(service.ShouldRetryGeo("agt_1"));
+        time.Advance(TimeSpan.FromMinutes(11));
+        Assert.True(service.ShouldRetryGeo("agt_1"));
+
+        var found = new NodeInfoService(new MemoryStore(), new FakeGeo(new GeoInfo("jp", 1, "Example", false)), time);
+        await found.ObserveAddressAsync("agt_2", NodeKinds.Agent, "8.8.8.8", CancellationToken.None);
+        time.Advance(TimeSpan.FromMinutes(11));
+        Assert.False(found.ShouldRetryGeo("agt_2"));
+    }
+
     [Fact]
     public async Task Service_PlatformIsSavedAndSurvivesARestart()
     {