AWS EC2上.NET 8应用无法连接Vertex AI，.NET 4.8可正常运行

我使用Google.Cloud.AIPlatform.V1调用Vertex AI获取预测响应时，遇到连接超时错误，但相同代码在.NET 4.8控制台应用、本地或私有虚拟机环境中可正常运行，仅在AWS EC2环境中出现问题。已尝试设置http_proxy和grpc_proxy，问题依旧。

错误信息 #

Grpc.Core.RpcException: Status(StatusCode="Unavailable", Detail="Error connecting to subchannel.", DebugException="System.Net.Sockets.SocketException: A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.")

System.Net.Sockets.SocketException (10060): A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

at System.Net.Sockets.Socket.AwaitableSocketAsyncEventArgs.System.Threading.Tasks.Sources.IValueTaskSource.GetResult(Int16 token)
at System.Net.Sockets.Socket.g__WaitForConnectWithCancellation|285_0(AwaitableSocketAsyncEventArgs saea, ValueTask connectTask, CancellationToken cancellationToken)
at Grpc.Net.Client.Balancer.Internal.SocketConnectivitySubchannelTransport.TryConnectAsync(ConnectContext context)
--- End of inner exception stack trace ---
at Grpc.Net.Client.Balancer.Internal.ConnectionManager.PickAsync(PickContext context, Boolean waitForReady, CancellationToken cancellationToken)
at Grpc.Net.Client.Balancer.Internal.BalancerHttpHandler.SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
at Grpc.Net.Client.Internal.GrpcCall2.RunCall(HttpRequestMessage request, Nullable1 timeout)
at Google.Api.Gax.Grpc.ApiCallRetryExtensions.<>c__DisplayClass0_0`2.<b__0>d.MoveNext()

--- End of stack trace from previous location ---

调用代码 #

PredictionServiceClientBuilder ClientBuilder = new PredictionServiceClientBuilder();
ClientBuilder.JsonCredentials = File.ReadAllText(credentialsPath);
ClientBuilder.Endpoint = "us-central1-aiplatform.googleapis.com";

var _predictionServiceClient = ClientBuilder.Build();

Console.WriteLine("Asking Google Vertex - How are you");
var prompt = "How are you ?";
var endpoint = EndpointName.FromProjectLocationPublisherModel("xxxx-xxxx-xxxx", "us-central1", "google", "text-bison@001").ToString();
var content = prompt;
var instanceValue = new wkt::Value
{
    StructValue = new wkt::Struct
    {
        Fields =
        {
            {
                "content", new wkt::Value
                {
                    StringValue = content
                }
            }
        }
    }
};

var instances = new List<wkt::Value>
{
    instanceValue
};

var parameters = new wkt::Value
{
    StructValue = new wkt::Struct
    {
        Fields =
        {
            { "temperature", new wkt::Value { NumberValue = 0.5} },
            { "maxOutputTokens", new wkt::Value { NumberValue = 1000 } },
            { "topP", new wkt::Value { NumberValue = 0.45 } },
            { "topK", new wkt::Value { NumberValue = 10 } }
        }
    }
};

var response = await _predictionServiceClient.PredictAsync(endpoint, instances, parameters);
Console.WriteLine();

Console.WriteLine("Response - ");
Console.WriteLine(response.Predictions[0].StructValue.Fields["content"].StringValue);

解决建议 #

• 检查EC2网络出站规则：确认EC2实例安全组出站规则允许访问us-central1-aiplatform.googleapis.com的443端口（HTTPS）；若使用VPC，需确保NAT网关/互联网网关正常工作，或配置了Vertex AI的VPC端点。
• 验证代理配置有效性：AWS环境的代理可能需要身份验证，或地址、端口配置错误。在EC2实例上用curl https://us-central1-aiplatform.googleapis.com测试代理连通性，同时确保GRPC代理支持HTTP/2协议。
• 统一.NET环境版本：确保EC2上使用的.NET版本与本地（.NET 4.8）一致，避免GRPC客户端兼容性问题。
• 添加超时与重试策略：在客户端构建时显式设置超时和重试逻辑，应对网络波动：

  ClientBuilder.CallSettings = CallSettings.FromRetry(RetrySettings.FromExponentialBackoff(
      maxAttempts: 5,
      initialBackoff: TimeSpan.FromSeconds(1),
      maxBackoff: TimeSpan.FromSeconds(10),
      backoffMultiplier: 2
  ).WithRetryFilter(status => status.StatusCode == StatusCode.Unavailable));
  

• 检查DNS解析：在EC2实例执行nslookup us-central1-aiplatform.googleapis.com，确认域名能正常解析；若失败，检查VPC的DNS设置是否启用了解析服务。

内容的提问来源于stack exchange，提问作者Nitin A