OA2023/OASystem/OASystem.Api/OAMethodLib/TokenHubAI/TokenHubService.cs

using Microsoft.EntityFrameworkCore.Internal;
using Microsoft.Extensions.Logging.Abstractions;
using OASystem.API.OAMethodLib.TokenHubAI.Models;
using System.Net;
using System.Text.Json;

namespace OASystem.API.OAMethodLib.TokenHubAI;

/// <summary>
/// TokenHub AI 服务实现
/// </summary>
public class TokenHubService : ITokenHubService, IDisposable
{
    private readonly HttpClient _httpClient;
    private readonly string _apiKey;
    private readonly string _baseUrl;
    private readonly JsonSerializerOptions _jsonOptions;
    private readonly ILogger<TokenHubService> _logger;

    // 重试配置
    private int _maxRetries = 3;
    private TimeSpan _initialRetryDelay = TimeSpan.FromSeconds(1);
    private TimeSpan _maxRetryDelay = TimeSpan.FromSeconds(30);

    public TokenHubService(
        string apiKey,
        string? baseUrl = null,
        ILogger<TokenHubService>? logger = null)
    {
        if (string.IsNullOrWhiteSpace(apiKey))
            throw new ArgumentException("API Key 不能为空", nameof(apiKey));

        _apiKey = apiKey;
        _baseUrl = baseUrl ?? "https://tokenhub.tencentmaas.com/v1/chat/completions";
        _logger = logger ?? NullLogger<TokenHubService>.Instance;

        _httpClient = new HttpClient();
        _httpClient.DefaultRequestHeaders.Add("Authorization", $"Bearer {_apiKey}");
        _httpClient.DefaultRequestHeaders.Add("Accept", "application/json");
        _httpClient.Timeout = TimeSpan.FromSeconds(120);

        _jsonOptions = new JsonSerializerOptions
        {
            PropertyNamingPolicy = JsonNamingPolicy.CamelCase,
            DefaultIgnoreCondition = System.Text.Json.Serialization.JsonIgnoreCondition.WhenWritingNull
        };
    }

    /// <summary>
    /// 发送聊天请求（非流式）
    /// </summary>
    public async Task<ChatResponse> ChatAsync(ChatRequest request, CancellationToken cancellationToken = default)
    {
        try
        {
            return await SendChatRequestAsync(request, cancellationToken);
        }
        catch (TokenHubException ex)
        {
            _logger.LogWarning(ex, ex.ErrorDetail?.Message ?? "-");

            var message = ex.ErrorDetail?.Message_ZH
                ?? TokenHubErrorExtensions.GetMessage(ex.ErrorCode);

            throw new TokenHubException(message, 400, ex.ErrorDetail);
        }
        catch (TaskCanceledException ex) when (ex.CancellationToken == cancellationToken)
        {
            _logger.LogWarning(ex, "请求被取消");
            throw new TokenHubException("请求已被取消", 499, new TokenHubError
            {
                Type = "server_error",
                Code = "499001",
                Message = "请求已被客户端取消，请检查客户端连接状态"
            });
        }
        catch (TaskCanceledException ex)
        {
            _logger.LogWarning(ex, "请求超时");
            throw new TokenHubException("请求超时", 504, new TokenHubError
            {
                Type = "timeout_error",
                Code = "504001",
                Message = "服务响应超时，请稍后重试"
            });
        }
        catch (Exception ex) when (ex is not TokenHubException)
        {
            _logger.LogError(ex, "发送请求时发生未预期错误");
            throw new TokenHubException($"请求失败: {ex.Message}", 500, new TokenHubError
            {
                Type = "server_error",
                Code = "500001",
                Message = "服务内部发生错误，请稍后重试或联系技术支持"
            });
        }
    }

    /// <summary>
    /// 发送聊天请求（流式）
    /// </summary>
    public async Task StreamChatAsync(ChatRequest request, Action<StreamChunk> onChunk, CancellationToken cancellationToken = default)
    {
        if (onChunk == null)
            throw new ArgumentNullException(nameof(onChunk));

        _httpClient.DefaultRequestHeaders.Remove("Accept");
        _httpClient.DefaultRequestHeaders.Add("Accept", "text/event-stream");

        try
        {
            await SendStreamRequestAsync(request, onChunk, cancellationToken);
        }
        catch (TokenHubException ex)
        {
            var message = ex.ErrorDetail?.Message_ZH
                ?? TokenHubErrorExtensions.GetMessage(ex.ErrorCode);

            throw new TokenHubException(message, 400, ex.ErrorDetail);
        }
        catch (TaskCanceledException ex) when (ex.CancellationToken == cancellationToken)
        {
            _logger.LogWarning(ex, "流式请求被取消");
            throw new TokenHubException("请求已被取消", 499, new TokenHubError
            {
                Type = "server_error",
                Code = "499001",
                Message = "request was canceled"
            });
        }
        catch (TaskCanceledException ex)
        {
            _logger.LogWarning(ex, "流式请求超时");
            throw new TokenHubException("请求超时", 504, new TokenHubError
            {
                Type = "timeout_error",
                Code = "504001",
                Message = "gateway timeout"
            });
        }
        catch (Exception ex) when (ex is not TokenHubException)
        {
            _logger.LogError(ex, "发送流式请求时发生未预期错误");
            throw new TokenHubException($"请求失败: {ex.Message}", 500, new TokenHubError
            {
                Type = "server_error",
                Code = "500001",
                Message = ex.Message
            });
        }
        finally
        {
            _httpClient.DefaultRequestHeaders.Remove("Accept");
            _httpClient.DefaultRequestHeaders.Add("Accept", "application/json");
        }
    }

    /// <summary>
    /// 带重试的聊天请求
    /// </summary>
    public async Task<ChatResponse> ChatWithRetryAsync(ChatRequest request, int maxRetries = 3, CancellationToken cancellationToken = default)
    {
        var retryCount = 0;
        var delay = _initialRetryDelay;

        while (true)
        {
            try
            {
                return await ChatAsync(request, cancellationToken);
            }
            catch (TokenHubException ex) when (ex.IsRetryable && retryCount < maxRetries)
            {
                retryCount++;
                _logger.LogWarning(ex,
                    "请求失败 (尝试 {RetryCount}/{MaxRetries})，将在 {DelayMs}ms 后重试",
                    retryCount, maxRetries, delay.TotalMilliseconds);

                await Task.Delay(delay, cancellationToken);

                // 指数退避
                delay = TimeSpan.FromSeconds(Math.Min(
                    delay.TotalSeconds * 2,
                    _maxRetryDelay.TotalSeconds
                ));
            }
            catch (TokenHubException ex) when (ex.ErrorCode == "429001" || ex.ErrorCode == "429002")
            {
                // 限流错误，使用更长的退避时间
                retryCount++;
                var retryAfter = TimeSpan.FromSeconds(Math.Min(
                    delay.TotalSeconds * 3,
                    _maxRetryDelay.TotalSeconds
                ));

                _logger.LogWarning(ex,
                    "触发限流 (尝试 {RetryCount}/{MaxRetries})，将在 {DelayMs}ms 后重试",
                    retryCount, maxRetries, retryAfter.TotalMilliseconds);

                await Task.Delay(retryAfter, cancellationToken);
                delay = retryAfter;
            }
            catch (TokenHubException)
            {
                // 不可重试的错误直接抛出
                throw;
            }
        }
    }

    /// <summary>
    /// 获取可用模型列表
    /// </summary>
    public async Task<List<string>> GetAvailableModelsAsync(CancellationToken cancellationToken = default)
    {
        // TokenHub 当前没有公开的模型列表 API
        return await Task.FromResult(new List<string>
            {
                "hunyuan-2.0-instruct-20251111"
            });
    }

    /// <summary>
    /// 设置超时时间
    /// </summary>
    public void SetTimeout(TimeSpan timeout)
    {
        _httpClient.Timeout = timeout;
    }

    /// <summary>
    /// 设置重试策略
    /// </summary>
    public void SetRetryPolicy(int maxRetries, TimeSpan? initialDelay = null, TimeSpan? maxDelay = null)
    {
        _maxRetries = maxRetries;
        if (initialDelay.HasValue)
            _initialRetryDelay = initialDelay.Value;
        if (maxDelay.HasValue)
            _maxRetryDelay = maxDelay.Value;
    }

    #region 私有方法

    /// <summary>
    /// 发送非流式请求
    /// </summary>
    private async Task<ChatResponse> SendChatRequestAsync(ChatRequest request, CancellationToken cancellationToken)
    {
        var requestBody = BuildRequestBody(request, stream: false);
        var json = System.Text.Json.JsonSerializer.Serialize(requestBody, _jsonOptions);
        var content = new StringContent(json, Encoding.UTF8, "application/json");

        _logger.LogDebug("发送请求到 TokenHub: Model={Model}, Messages={MessageCount}",
            request.Model, request.Messages.Count);

        var httpResponse = await _httpClient.PostAsync(_baseUrl, content, cancellationToken);
        var responseJson = await httpResponse.Content.ReadAsStringAsync(cancellationToken);

        _logger.LogDebug("收到响应: StatusCode={StatusCode}, RequestId={RequestId}",
            httpResponse.StatusCode,
            ExtractRequestId(responseJson));

        if (!httpResponse.IsSuccessStatusCode)
        {
            throw CreateExceptionFromResponse(httpResponse.StatusCode, responseJson);
        }

        return ParseChatResponse(responseJson);
    }

    /// <summary>
    /// 发送流式请求
    /// </summary>
    private async Task SendStreamRequestAsync(
        ChatRequest request,
        Action<StreamChunk> onChunk,
        CancellationToken cancellationToken)
    {
        var requestBody = BuildRequestBody(request, stream: true);
        var json = System.Text.Json.JsonSerializer.Serialize(requestBody, _jsonOptions);
        var content = new StringContent(json, Encoding.UTF8, "application/json");

        _logger.LogDebug("发送流式请求到 TokenHub: Model={Model}", request.Model);

        var httpResponse = await _httpClient.PostAsync(_baseUrl, content, cancellationToken);

        if (!httpResponse.IsSuccessStatusCode)
        {
            var errorJson = await httpResponse.Content.ReadAsStringAsync(cancellationToken);
            throw CreateExceptionFromResponse(httpResponse.StatusCode, errorJson);
        }

        using var stream = await httpResponse.Content.ReadAsStreamAsync(cancellationToken);
        using var reader = new StreamReader(stream);

        string? line;
        while ((line = await reader.ReadLineAsync()) != null)
        {
            cancellationToken.ThrowIfCancellationRequested();

            if (string.IsNullOrWhiteSpace(line) || !line.StartsWith("data: "))
                continue;

            var data = line.Substring(6);

            if (data == "[DONE]")
            {
                onChunk(new StreamChunk { IsDone = true });
                break;
            }

            try
            {
                var chunk = ParseStreamChunk(data);
                if (chunk != null)
                {
                    onChunk(chunk);
                    if (chunk.IsDone)
                        break;
                }
            }
            catch (System.Text.Json.JsonException ex)
            {
                _logger.LogWarning(ex, "解析流式数据失败: {Data}", data);
                // 跳过无法解析的数据
            }
        }
    }

    /// <summary>
    /// 构建请求体
    /// </summary>
    private object BuildRequestBody(ChatRequest request, bool stream)
    {
        var messages = new List<object>();

        if (!string.IsNullOrEmpty(request.SystemPrompt))
        {
            messages.Add(new { role = "system", content = request.SystemPrompt });
        }

        foreach (var msg in request.Messages)
        {
            messages.Add(new { role = msg.Role, content = msg.Content });
        }

        return new
        {
            model = request.Model,
            messages = messages,
            max_tokens = request.MaxTokens,
            temperature = request.Temperature,
            stream = stream
        };
    }

    /// <summary>
    /// 解析聊天响应
    /// </summary>
    private ChatResponse ParseChatResponse(string json)
    {
        using var document = JsonDocument.Parse(json);
        var root = document.RootElement;

        var response = new ChatResponse
        {
            RawJson = json,
            IsSuccess = true
        };

        if (root.TryGetProperty("id", out var id))
            response.Id = id.GetString();

        if (root.TryGetProperty("model", out var model))
            response.Model = model.GetString();

        if (root.TryGetProperty("choices", out var choices) && choices.GetArrayLength() > 0)
        {
            var choice = choices[0];
            if (choice.TryGetProperty("message", out var message))
            {
                if (message.TryGetProperty("content", out var content))
                    response.Content = content.GetString();
            }

            if (choice.TryGetProperty("finish_reason", out var finishReason))
            {
                // 检查是否被内容安全过滤
                if (finishReason.GetString() == "content_filter")
                {
                    response.IsSuccess = false;
                    response.ErrorMessage = "内容被安全策略过滤";
                }
            }
        }

        if (root.TryGetProperty("usage", out var usage))
        {
            response.Usage = new TokenUsage
            {
                PromptTokens = usage.TryGetProperty("prompt_tokens", out var pt) ? pt.GetInt32() : 0,
                CompletionTokens = usage.TryGetProperty("completion_tokens", out var ct) ? ct.GetInt32() : 0,
                TotalTokens = usage.TryGetProperty("total_tokens", out var tt) ? tt.GetInt32() : 0
            };
        }

        return response;
    }

    /// <summary>
    /// 解析流式数据块
    /// </summary>
    private StreamChunk? ParseStreamChunk(string data)
    {
        using var document = JsonDocument.Parse(data);
        var root = document.RootElement;

        var chunk = new StreamChunk();

        if (root.TryGetProperty("model", out var model))
            chunk.Model = model.GetString();

        if (root.TryGetProperty("choices", out var choices) && choices.GetArrayLength() > 0)
        {
            var choice = choices[0];

            if (choice.TryGetProperty("delta", out var delta))
            {
                if (delta.TryGetProperty("content", out var content))
                {
                    chunk.Content = content.GetString() ?? string.Empty;
                }
            }

            // 检查是否结束
            if (choice.TryGetProperty("finish_reason", out var finishReason))
            {
                var reason = finishReason.GetString();
                chunk.IsDone = reason == "stop" || reason == "content_filter";

                if (reason == "content_filter")
                {
                    chunk.Content = "[内容被安全策略过滤]";
                }
            }
        }

        return chunk;
    }

    /// <summary>
    /// 从错误响应创建异常
    /// </summary>
    private TokenHubException CreateExceptionFromResponse(HttpStatusCode statusCode, string responseJson)
    {
        var statusCodeInt = (int)statusCode;
        var errorDetail = ParseErrorResponse(responseJson, statusCodeInt);

        var message = errorDetail?.Message ?? $"HTTP {(int)statusCode}: 请求失败";

        _logger.LogWarning("TokenHub API 返回错误: StatusCode={StatusCode}, Code={Code}, Type={Type}, RequestId={RequestId}",
            statusCodeInt,
            errorDetail?.Code,
            errorDetail?.Type,
            errorDetail?.RequestId);

        return new TokenHubException(message, statusCodeInt, errorDetail);
    }

    /// <summary>
    /// 解析错误响应
    /// </summary>
    private TokenHubError? ParseErrorResponse(string responseJson, int statusCode)
    {
        try
        {
            using var document = JsonDocument.Parse(responseJson);
            var root = document.RootElement;

            // 判断是 OpenAI 协议还是 Anthropic 协议
            if (root.TryGetProperty("error", out var errorElement))
            {
                // OpenAI 协议
                if (errorElement.ValueKind == JsonValueKind.Object)
                {
                    return ParseOpenAiError(errorElement);
                }
            }

            // Anthropic 协议
            if (root.TryGetProperty("type", out var typeProp) && typeProp.GetString() == "error")
            {
                if (root.TryGetProperty("error", out var anthropicError))
                {
                    return ParseAnthropicError(anthropicError);
                }
            }

            // 简化格式（链路短路等）
            if (root.TryGetProperty("error", out var simpleError) &&
                simpleError.ValueKind == JsonValueKind.Object)
            {
                return new TokenHubError
                {
                    Type = simpleError.TryGetProperty("type", out var t) ? t.GetString() : null,
                    Code = simpleError.TryGetProperty("code", out var c) ? c.GetString() : null,
                    Message = simpleError.TryGetProperty("message", out var m) ? m.GetString() : null,
                    Message_ZH = simpleError.TryGetProperty("message_zh", out var m_zh) ? m_zh.GetString() : null,
                    RequestId = simpleError.TryGetProperty("request_id", out var r) ? r.GetString() : null
                };
            }

            return null;
        }
        catch (System.Text.Json.JsonException ex)
        {
            _logger.LogWarning(ex, "解析错误响应失败: {ResponseJson}", responseJson);
            return null;
        }
    }

    /// <summary>
    /// 解析 OpenAI 协议错误
    /// </summary>
    private TokenHubError ParseOpenAiError(JsonElement errorElement)
    {
        var error = new TokenHubError();

        if (errorElement.TryGetProperty("type", out var type))
            error.Type = type.GetString();

        if (errorElement.TryGetProperty("code", out var code))
            error.Code = code.GetString();

        if (errorElement.TryGetProperty("message", out var message))
            error.Message = message.GetString();

        if (errorElement.TryGetProperty("message_zh", out var message_zh))
            error.Message_ZH = message_zh.GetString();

        if (errorElement.TryGetProperty("source", out var source))
            error.Source = source.GetString();

        if (errorElement.TryGetProperty("upstream_code", out var upstreamCode))
            error.UpstreamCode = upstreamCode.GetString();

        if (errorElement.TryGetProperty("upstream_status", out var upstreamStatus))
            error.UpstreamStatus = upstreamStatus.GetInt32();

        if (errorElement.TryGetProperty("request_id", out var requestId))
            error.RequestId = requestId.GetString();

        return error;
    }

    /// <summary>
    /// 解析 Anthropic 协议错误
    /// </summary>
    private TokenHubError ParseAnthropicError(JsonElement errorElement)
    {
        var error = new TokenHubError
        {
            Source = "upstream" // Anthropic 协议默认视为上游
        };

        if (errorElement.TryGetProperty("type", out var type))
            error.Type = type.GetString();

        if (errorElement.TryGetProperty("message", out var message))
            error.Message = message.GetString();

        if (errorElement.TryGetProperty("message_zh", out var message_zh))
            error.Message_ZH = message_zh.GetString();

        if (errorElement.TryGetProperty("reqid", out var reqid))
            error.RequestId = reqid.GetString();

        // Anthropic 协议没有业务码，根据类型映射
        error.Code = MapAnthropicTypeToCode(error.Type);

        return error;
    }

    /// <summary>
    /// 将 Anthropic 错误类型映射到业务码
    /// </summary>
    private string? MapAnthropicTypeToCode(string? type)
    {
        return type switch
        {
            "invalid_request_error" => "400002",
            "authentication_error" => "401002",
            "permission_error" => "403001",
            "not_found_error" => "400004",
            "rate_limit_error" => "429001",
            "overloaded_error" => "502001",
            "api_error" => "500001",
            _ => null
        };
    }

    /// <summary>
    /// 从响应中提取 Request ID
    /// </summary>
    private string? ExtractRequestId(string responseJson)
    {
        try
        {
            using var document = JsonDocument.Parse(responseJson);
            var root = document.RootElement;

            if (root.TryGetProperty("error", out var error))
            {
                if (error.TryGetProperty("request_id", out var requestId))
                    return requestId.GetString();
            }

            if (root.TryGetProperty("id", out var id))
                return id.GetString();
        }
        catch
        {
            // 忽略解析错误
        }

        return null;
    }

    #endregion

    /// <summary>
    /// 释放资源
    /// </summary>
    public void Dispose()
    {
        _httpClient?.Dispose();
        GC.SuppressFinalize(this);
    }
}