feat(kb): VikingDB纯检索+重排+Redis上下文+全库搜索+别名扩展+KB保护窗口+RAG语气引导

- 新增 kbRetriever.js: VikingDB search_knowledge 纯检索替代 Ark chat/completions, doubao-seed-rerank 重排, RAG payload 语气引导缓解音色差异 - 新增 redisClient.js: Redis 连接管理 + 5轮对话历史 + KB缓存双写 - toolExecutor.js: 产品别名扩展25条, 全库检索topK=25, 检索阈值0.01, 精简 buildDeterministicKnowledgeQuery - nativeVoiceGateway.js: isPureChitchat扩展, KB保护窗口60s, prequery参数调优 - realtimeDialogRouting.js: resolveReply感知KB保护窗口, fast-path适配raw模式 - app.js: 健康检查新增 redis/reranker/kbRetrievalMode - 新增测试: alias A/B测试, KB retriever测试, Redis客户端测试, raw模式集成测试
2026-03-26 14:30:32 +08:00
parent 9567eb7358
commit 56940676f6
15 changed files with 2096 additions and 170 deletions
--- a/test2/server/tests/test_raw_mode_integration.js
+++ b/test2/server/tests/test_raw_mode_integration.js
@@ -0,0 +1,331 @@
+/**
+ * raw 模式集成测试
+ * 覆盖：resolveReply raw/answer 模式切换、ragItems 格式、字幕过滤、缓存双写
+ *
+ * 运行方式: node --test tests/test_raw_mode_integration.js
+ */
+const { describe, it, beforeEach, afterEach } = require('node:test');
+const assert = require('node:assert/strict');
+const path = require('path');
+const fs = require('fs');
+
+// 加载 .env
+const envPath = path.join(__dirname, '../.env');
+if (fs.existsSync(envPath)) {
+  fs.readFileSync(envPath, 'utf8').split('\n').forEach(line => {
+    const trimmed = line.trim();
+    if (!trimmed || trimmed.startsWith('#')) return;
+    const idx = trimmed.indexOf('=');
+    if (idx > 0) {
+      const key = trimmed.slice(0, idx).trim();
+      let val = trimmed.slice(idx + 1).trim();
+      if ((val.startsWith('"') && val.endsWith('"')) || (val.startsWith("'") && val.endsWith("'"))) {
+        val = val.slice(1, -1);
+      }
+      if (!process.env[key]) process.env[key] = val;
+    }
+  });
+}
+
+const { normalizeTextForSpeech } = require('../services/realtimeDialogRouting');
+
+// ================================================================
+// 1. ragItems 格式验证 — raw 模式 vs answer 模式
+// ================================================================
+describe('ragItems 格式 — raw 模式 vs answer 模式', () => {
+
+  it('raw 模式 toolResult 应包含 retrieval_mode 标记', () => {
+    const rawToolResult = {
+      query: '小红怎么吃',
+      results: [
+        { title: '对话上下文', content: '用户: 你好\n助手: 你好！' },
+        { title: '产品手册', content: '小红Activize Oxyplus...' },
+        { title: 'FAQ', content: '小红每天一包...' },
+        { title: '知识库片段3', content: '用40度以下温水冲服...' },
+      ],
+      total: 4,
+      source: 'ark_knowledge',
+      hit: true,
+      reason: 'reranked_hit',
+      retrieval_mode: 'raw',
+      top_score: 0.85,
+      chunks_count: 3,
+    };
+
+    assert.equal(rawToolResult.retrieval_mode, 'raw');
+    assert.equal(rawToolResult.results.length, 4, 'Should have context + 3 chunks');
+    assert.equal(rawToolResult.results[0].title, '对话上下文', 'First item should be context');
+  });
+
+  it('answer 模式 toolResult 不应包含 retrieval_mode 或为 undefined', () => {
+    const answerToolResult = {
+      query: '小红怎么吃',
+      results: [{ title: '方舟知识库检索结果', content: '根据知识库信息，小红Activize每天一包...' }],
+      total: 1,
+      source: 'ark_knowledge',
+      hit: true,
+      reason: 'classified_hit',
+    };
+
+    assert.ok(!answerToolResult.retrieval_mode || answerToolResult.retrieval_mode !== 'raw');
+  });
+
+  it('raw 模式下 ragItems 应透传多条而非合并为单条', () => {
+    const rawResults = [
+      { title: '对话上下文', content: '用户: 大白怎么吃\n助手: ...' },
+      { title: '产品手册', content: '片段1内容' },
+      { title: 'FAQ', content: '片段2内容' },
+    ];
+
+    const isRawMode = true;
+    let finalRagItems;
+    if (isRawMode) {
+      finalRagItems = rawResults;
+    } else {
+      const replyText = rawResults.map(r => r.content).join(' ');
+      finalRagItems = [{ title: '知识库结果', content: replyText }];
+    }
+
+    assert.equal(finalRagItems.length, 3, 'Raw mode should keep 3 items');
+    assert.equal(finalRagItems[0].title, '对话上下文');
+    assert.equal(finalRagItems[1].title, '产品手册');
+  });
+
+  it('answer 模式下 ragItems 应合并为单条并清理前缀', () => {
+    const replyText = '根据知识库信息，小红Activize每天一包冲服';
+    const isRawMode = false;
+
+    let finalRagItems;
+    if (!isRawMode) {
+      const speechText = normalizeTextForSpeech(replyText);
+      const cleanedText = speechText.replace(/^(根据知识库信息[，,：:\s]*|根据.*?[，,]\s*)/i, '');
+      finalRagItems = [{ title: '知识库结果', content: cleanedText || speechText }];
+    }
+
+    assert.equal(finalRagItems.length, 1, 'Answer mode should merge into 1 item');
+    assert.ok(!finalRagItems[0].content.startsWith('根据知识库信息'), 'Should strip prefix');
+    assert.ok(finalRagItems[0].content.includes('小红Activize'), 'Should preserve core content');
+  });
+});
+
+// ================================================================
+// 2. 字幕过滤 — raw 模式排除"对话上下文"
+// ================================================================
+describe('字幕过滤 — raw 模式排除对话上下文', () => {
+
+  it('raw 模式的 ragContent 过滤后字幕不应包含上下文', () => {
+    const ragContent = [
+      { title: '对话上下文', content: '用户: 你好\n助手: 你好' },
+      { title: '产品手册', content: '小红每天一包' },
+      { title: 'FAQ', content: '40度以下温水' },
+    ];
+
+    const subtitleItems = ragContent.filter((item) => item.title !== '对话上下文');
+    const ragSubtitleText = (subtitleItems.length > 0 ? subtitleItems : ragContent)
+      .map((item) => item.content).join(' ');
+
+    assert.ok(!ragSubtitleText.includes('用户: 你好'), 'Subtitle should not include context');
+    assert.ok(ragSubtitleText.includes('小红每天一包'), 'Subtitle should include KB content');
+    assert.ok(ragSubtitleText.includes('40度以下温水'), 'Subtitle should include all KB chunks');
+  });
+
+  it('answer 模式无"对话上下文"条目，字幕应正常', () => {
+    const ragContent = [
+      { title: '知识库结果', content: '小红Activize每天一包冲服' },
+    ];
+
+    const subtitleItems = ragContent.filter((item) => item.title !== '对话上下文');
+    const ragSubtitleText = (subtitleItems.length > 0 ? subtitleItems : ragContent)
+      .map((item) => item.content).join(' ');
+
+    assert.equal(ragSubtitleText, '小红Activize每天一包冲服');
+  });
+
+  it('全部都是上下文条目时应降级显示全部（兜底）', () => {
+    const ragContent = [
+      { title: '对话上下文', content: '用户: 测试' },
+    ];
+
+    const subtitleItems = ragContent.filter((item) => item.title !== '对话上下文');
+    const ragSubtitleText = (subtitleItems.length > 0 ? subtitleItems : ragContent)
+      .map((item) => item.content).join(' ');
+
+    assert.equal(ragSubtitleText, '用户: 测试', 'Should fallback to showing context');
+  });
+});
+
+// ================================================================
+// 3. 环境变量模式切换
+// ================================================================
+describe('环境变量模式切换', () => {
+  const ENV_ORIG = {};
+
+  beforeEach(() => {
+    ENV_ORIG.mode = process.env.VOLC_ARK_KB_RETRIEVAL_MODE;
+    ENV_ORIG.reranker = process.env.ENABLE_RERANKER;
+    ENV_ORIG.redis = process.env.ENABLE_REDIS_CONTEXT;
+  });
+
+  afterEach(() => {
+    if (ENV_ORIG.mode !== undefined) process.env.VOLC_ARK_KB_RETRIEVAL_MODE = ENV_ORIG.mode;
+    else delete process.env.VOLC_ARK_KB_RETRIEVAL_MODE;
+    if (ENV_ORIG.reranker !== undefined) process.env.ENABLE_RERANKER = ENV_ORIG.reranker;
+    else delete process.env.ENABLE_RERANKER;
+    if (ENV_ORIG.redis !== undefined) process.env.ENABLE_REDIS_CONTEXT = ENV_ORIG.redis;
+    else delete process.env.ENABLE_REDIS_CONTEXT;
+  });
+
+  it('VOLC_ARK_KB_RETRIEVAL_MODE=raw 应使用新链路', () => {
+    process.env.VOLC_ARK_KB_RETRIEVAL_MODE = 'raw';
+    const mode = process.env.VOLC_ARK_KB_RETRIEVAL_MODE || 'answer';
+    assert.equal(mode, 'raw');
+  });
+
+  it('VOLC_ARK_KB_RETRIEVAL_MODE=answer 应使用旧链路', () => {
+    process.env.VOLC_ARK_KB_RETRIEVAL_MODE = 'answer';
+    const mode = process.env.VOLC_ARK_KB_RETRIEVAL_MODE || 'answer';
+    assert.equal(mode, 'answer');
+  });
+
+  it('VOLC_ARK_KB_RETRIEVAL_MODE 未设置时默认 answer', () => {
+    delete process.env.VOLC_ARK_KB_RETRIEVAL_MODE;
+    const mode = process.env.VOLC_ARK_KB_RETRIEVAL_MODE || 'answer';
+    assert.equal(mode, 'answer');
+  });
+
+  it('ENABLE_RERANKER=false 应关闭重排', () => {
+    process.env.ENABLE_RERANKER = 'false';
+    assert.equal(process.env.ENABLE_RERANKER !== 'false', false);
+  });
+
+  it('ENABLE_REDIS_CONTEXT=false 应关闭 Redis 上下文', () => {
+    process.env.ENABLE_REDIS_CONTEXT = 'false';
+    assert.equal(process.env.ENABLE_REDIS_CONTEXT !== 'false', false);
+  });
+
+  it('所有降级开关独立，互不影响', () => {
+    process.env.VOLC_ARK_KB_RETRIEVAL_MODE = 'raw';
+    process.env.ENABLE_RERANKER = 'false';
+    process.env.ENABLE_REDIS_CONTEXT = 'true';
+
+    assert.equal(process.env.VOLC_ARK_KB_RETRIEVAL_MODE, 'raw', 'mode should be raw');
+    assert.equal(process.env.ENABLE_RERANKER, 'false', 'reranker should be off');
+    assert.equal(process.env.ENABLE_REDIS_CONTEXT, 'true', 'redis context should be on');
+  });
+});
+
+// ================================================================
+// 4. normalizeTextForSpeech 对 raw 内容的处理
+// ================================================================
+describe('normalizeTextForSpeech — raw 片段处理', () => {
+
+  it('应去除 Markdown 标记', () => {
+    const input = '## 产品介绍\n**小红Activize** 是一款_细胞营养素_';
+    const result = normalizeTextForSpeech(input);
+    assert.ok(!result.includes('##'), 'Should remove heading marks');
+    assert.ok(!result.includes('**'), 'Should remove bold marks');
+    assert.ok(!result.includes('_细胞'), 'Should remove italic marks');
+    assert.ok(result.includes('小红Activize'), 'Should preserve content');
+  });
+
+  it('纯文本内容不应被破坏', () => {
+    const input = '小红Activize Oxyplus每天一包，用40度以下温水冲服，搅拌均匀后饮用。';
+    const result = normalizeTextForSpeech(input);
+    assert.equal(result, input, 'Plain text should not be modified');
+  });
+
+  it('空字符串应返回空', () => {
+    assert.equal(normalizeTextForSpeech(''), '');
+    assert.equal(normalizeTextForSpeech(null), '');
+    assert.equal(normalizeTextForSpeech(undefined), '');
+  });
+});
+
+// ================================================================
+// 5. 端到端数据流验证
+// ================================================================
+describe('端到端数据流 — raw 模式完整 ragPayload → S2S', () => {
+
+  it('模拟完整 raw 模式数据流：query → chunks → rerank → payload → ragItems', () => {
+    // Step 1: 模拟检索结果
+    const retrievedChunks = [
+      { id: 'c1', content: '小红Activize Oxyplus是德国PM的核心产品', score: 0.9, doc_name: '产品手册' },
+      { id: 'c2', content: '每天一包，40度以下温水冲服', score: 0.85, doc_name: '使用说明' },
+      { id: 'c3', content: '含有瓜拉纳提取物，提供天然能量', score: 0.75, doc_name: '成分表' },
+    ];
+
+    // Step 2: 模拟重排（已排序，直接取 top3）
+    const rerankedChunks = retrievedChunks.slice(0, 3);
+
+    // Step 3: 模拟历史
+    const history = [
+      { role: 'user', content: '你们有什么产品' },
+      { role: 'assistant', content: '我们有基础三合一，包括大白小红小白...' },
+    ];
+
+    // Step 4: 构建 payload
+    const kbRetriever = require('../services/kbRetriever');
+    const ragPayload = kbRetriever.buildRagPayload(rerankedChunks, history);
+
+    // 验证 payload 结构
+    assert.equal(ragPayload.length, 4, 'Should be context(1) + chunks(3)');
+    assert.equal(ragPayload[0].title, '对话上下文');
+    assert.ok(ragPayload[0].content.includes('你们有什么产品'));
+    assert.equal(ragPayload[1].title, '产品手册');
+    assert.equal(ragPayload[2].title, '使用说明');
+    assert.equal(ragPayload[3].title, '成分表');
+
+    // Step 5: 模拟 toolResult
+    const toolResult = {
+      results: ragPayload.map(item => ({ title: item.title, content: item.content })),
+      hit: true,
+      retrieval_mode: 'raw',
+    };
+
+    // Step 6: 模拟 resolveReply 中的 ragItems 构建
+    const ragItems = toolResult.results.filter(i => i && i.content).map(i => ({
+      title: i.title || '知识库结果',
+      content: i.content,
+    }));
+
+    const isRawMode = toolResult.retrieval_mode === 'raw';
+    const finalRagItems = isRawMode ? ragItems : [{ title: '知识库结果', content: ragItems.map(i => i.content).join(' ') }];
+
+    assert.equal(finalRagItems.length, 4, 'Raw mode: 4 items pass-through');
+
+    // Step 7: 模拟字幕过滤
+    const subtitleItems = finalRagItems.filter(item => item.title !== '对话上下文');
+    const subtitle = subtitleItems.map(item => item.content).join(' ');
+    assert.ok(!subtitle.includes('你们有什么产品'), 'Subtitle should exclude context');
+    assert.ok(subtitle.includes('小红Activize'), 'Subtitle should include chunk content');
+
+    // Step 8: 模拟 sendExternalRag 的 JSON 序列化
+    const jsonStr = JSON.stringify(finalRagItems);
+    const parsed = JSON.parse(jsonStr);
+    assert.equal(parsed.length, 4, 'JSON roundtrip should preserve all items');
+  });
+
+  it('模拟完整 answer 模式数据流（对比验证）', () => {
+    const toolResult = {
+      results: [{ title: '方舟知识库检索结果', content: '根据知识库信息，小红Activize每天一包冲服' }],
+      hit: true,
+    };
+
+    const ragItems = toolResult.results.filter(i => i && i.content).map(i => ({
+      title: i.title || '知识库结果',
+      content: i.content,
+    }));
+
+    const isRawMode = !!(toolResult.retrieval_mode === 'raw');
+    assert.equal(isRawMode, false, 'Should be answer mode');
+
+    const speechText = normalizeTextForSpeech(ragItems[0].content);
+    const cleanedText = speechText.replace(/^(根据知识库信息[，,：:\s]*|根据.*?[，,]\s*)/i, '');
+    const finalRagItems = [{ title: '知识库结果', content: cleanedText || speechText }];
+
+    assert.equal(finalRagItems.length, 1, 'Answer mode: single merged item');
+    assert.ok(!finalRagItems[0].content.startsWith('根据知识库'), 'Prefix should be stripped');
+  });
+});
+
+console.log('\n=== raw模式集成测试加载完成 ===\n');