build-flow.api/tests/BuildFlow.Rag.Tests/RagQAEndToEndTests.cs

using BuildFlow.Application.RagQA.Commands;
using BuildFlow.Application.RagQA.Handlers;
using BuildFlow.Domain.Entities.Document;
using BuildFlow.Domain.Entities.KnowledgeBase;
using BuildFlow.Domain.Enums;
using BuildFlow.Domain.Interfaces;
using BuildFlow.Infrastructure.Persistence;
using BuildFlow.Rag.Tests.Stub;
using FluentAssertions;
using Microsoft.EntityFrameworkCore;
using Xunit;

namespace BuildFlow.Rag.Tests;

/// <summary>
/// [SOP-RAG-01] 知识库 RAG 问答全链路端到端测试。
/// 用 Stub IRagRetrievalService 返回预设 sources,绕过 pgvector 真实召回。
/// 验证:RagQueryCommandHandler 取 KB → RetrieveAsync → IAIChatAgent.RunAsync → 返回 answer+sources。
/// 飞书文档: https://ecn1rid3e27z.feishu.cn/wiki/F9S2wBS77i35bkkwJyVccDlfnyg
/// </summary>
public class RagQAEndToEndTests
{
    private static async Task<RagDbContext> CreateDbAsync(string name)
    {
        var options = new DbContextOptionsBuilder<RagDbContext>()
            .UseInMemoryDatabase($"rag-qa-{name}-{Guid.NewGuid():N}")
            .Options;
        await using var db = new RagDbContext(options);
        return new RagDbContext(options);
    }

    [Fact]
    public async Task Query_WithSources_ReturnsAnswerAndSources()
    {
        // Given: 已创建知识库
        await using var db = await CreateDbAsync("with-sources");
        var kb = new KnowledgeBaseEntity
        {
            Name = "测试知识库",
            RetrievalMode = RetrievalModeEnum.Hybrid,
            RetrievalTopK = 5,
            ContextTopK = 3,
            SimilarityThreshold = 0.3,
            VectorWeight = 0.7,
            CreatedBy = "test",
            UpdatedBy = "test",
            OperatorIP = "127.0.0.1"
        };
        db.KnowledgeBases.Add(kb);

        var documentId = Guid.NewGuid();
        var chunkId = Guid.NewGuid();
        db.Documents.Add(new DocumentEntity
        {
            Id = documentId,
            KnowledgeBaseId = kb.Id,
            Title = "技术架构",
            FileName = "arch.md",
            FilePath = "test/arch.md",
            FileSize = 100,
            ContentType = "text/markdown",
            Status = DocumentStatusEnum.Completed,
            CreatedBy = "test",
            UpdatedBy = "test",
            OperatorIP = "127.0.0.1"
        });
        await db.SaveChangesAsync();

        // Stub RetrievalService 返回预设 sources
        var retrieval = new StubRagRetrievalService
        {
            Hits =
            [
                new RetrievalHit(
                    documentId,
                    "技术架构",
                    chunkId,
                    "BuildFlow 使用 PostgreSQL + pgvector",
                    0.87,
                    "BuildFlow 使用 PostgreSQL + pgvector")
            ]
        };
        var chatAgent = new StubAIChatAgent { ReplyText = "BuildFlow 使用 pgvector 向量数据库" };

        var handler = new RagQueryCommandHandler(db, retrieval, chatAgent);

        // When: 调用 RAG 查询
        var result = await handler.Handle(
            new RAGQueryCommand(kb.Id, "BuildFlow 用了什么向量数据库?"),
            CancellationToken.None);

        // Then: 返回 answer + sources
        result.Answer.Should().NotBeNullOrEmpty("应有 LLM 响应");
        result.Answer.Should().Be("BuildFlow 使用 pgvector 向量数据库");
        result.Sources.Should().NotBeEmpty("sources 非空");
        result.Sources[0].DocumentId.Should().Be(documentId, "DocumentId 匹配上传文档");
        result.Sources[0].Similarity.Should().BeGreaterThan(0, "相似度>0");
    }

    [Fact]
    public async Task Query_NoSources_ReturnsNotFoundMessage()
    {
        // Given: 知识库存在但 RetrievalService 返回空
        await using var db = await CreateDbAsync("no-sources");
        var kb = new KnowledgeBaseEntity
        {
            Name = "空知识库",
            CreatedBy = "test",
            UpdatedBy = "test",
            OperatorIP = "127.0.0.1"
        };
        db.KnowledgeBases.Add(kb);
        await db.SaveChangesAsync();

        var retrieval = new StubRagRetrievalService { Hits = [] };
        var chatAgent = new StubAIChatAgent();

        var handler = new RagQueryCommandHandler(db, retrieval, chatAgent);

        // When: 查询无结果
        var result = await handler.Handle(
            new RAGQueryCommand(kb.Id, "不存在的内容"),
            CancellationToken.None);

        // Then: 返回未找到提示,sources 空
        result.Sources.Should().BeEmpty("无召回结果");
        result.Answer.Should().NotBeNull("应返回提示文本");
    }

    [Theory]
    [InlineData(RetrievalModeEnum.Vector)]
    [InlineData(RetrievalModeEnum.Keyword)]
    [InlineData(RetrievalModeEnum.Hybrid)]
    public async Task Query_WorksAcrossRetrievalModes(RetrievalModeEnum mode)
    {
        // Given: 不同 RetrievalModeEnum 的知识库(Stub 不依赖真实召回算法)
        await using var db = await CreateDbAsync($"mode-{mode}");
        var kb = new KnowledgeBaseEntity
        {
            Name = $"模式测试-{mode}",
            RetrievalMode = mode,
            CreatedBy = "test",
            UpdatedBy = "test",
            OperatorIP = "127.0.0.1"
        };
        db.KnowledgeBases.Add(kb);
        await db.SaveChangesAsync();

        var documentId = Guid.NewGuid();
        var retrieval = new StubRagRetrievalService
        {
            Hits = [new RetrievalHit(documentId, "doc", Guid.NewGuid(), "内容", 0.9, "内容")]
        };
        var chatAgent = new StubAIChatAgent { ReplyText = "回答" };
        var handler = new RagQueryCommandHandler(db, retrieval, chatAgent);

        // When
        var result = await handler.Handle(new RAGQueryCommand(kb.Id, "问题"), CancellationToken.None);

        // Then: 三种模式都能返回结果(Stub 绕过真实算法差异)
        result.Sources.Should().HaveCount(1);
        result.Answer.Should().Be("回答");
    }
}