blandarebiter/bigwo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: conversation long-term memory + fix source ENUM bug

Browse Source 
- New: conversationSummarizer.js (LLM summary every 3 turns, loadBestSummary, persistFinalSummary)
- db/index.js: conversation_summaries table, upsertConversationSummary, getSessionSummary
- redisClient.js: setSummary/getSummary (TTL 2h)
- nativeVoiceGateway.js: _turnCount tracking, trigger summarize, persist on close
- realtimeDialogRouting.js: inject summary context, reduce history 5->3 rounds
- Fix: messages source ENUM missing 'search_knowledge' causing chat DB writes to fail
This commit is contained in:
User
2026-04-03 10:19:16 +08:00
parent 5b824cd16a
commit fe25229de7
6 changed files with 797 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

200
test2/server/services/conversationSummarizer.js Normal file
View File

@@ -0,0 +1,200 @@
const axios = require('axios');
const redisClient = require('./redisClient');
const db = require('../db');
// ============ 配置 ============
const ENABLED = (process.env.ENABLE_CONVERSATION_SUMMARY || 'true') !== 'false';
const SUMMARIZE_EVERY_N_TURNS = parseInt(process.env.SUMMARY_EVERY_N_TURNS) || 3;
const SUMMARY_MAX_TOKENS = parseInt(process.env.SUMMARY_MAX_TOKENS) || 120;
const MIN_TURNS_TO_PERSIST = parseInt(process.env.SUMMARY_MIN_TURNS_TO_PERSIST) || 2;
const SUMMARY_MODEL = process.env.VOLC_ARK_KB_MODEL || process.env.VOLC_ARK_ENDPOINT_ID || '';
const ARK_API_KEY = process.env.VOLC_ARK_API_KEY || process.env.VOLC_ACCESS_KEY_ID || '';
const ARK_BASE_URL = 'https://ark.cn-beijing.volces.com/api/v3/chat/completions';
const SUMMARY_PROMPT = `你是对话摘要助手。将以下对话历史浓缩为简短摘要，必须保留：
1. 用户询问过的所有产品名称和具体问题
2. AI给出的关键数字（剂量、价格、数量等）
3. 用户表达的偏好或关注点
4. 未解决的问题或用户的疑虑
规则：只输出摘要正文，不加前缀或标题。150字以内。用"用户"和"助手"指代双方。`;
// ============ LLM 摘要生成 ============
async function summarizeConversation(existingSummary, recentMessages) {
if (!ARK_API_KEY || !SUMMARY_MODEL) {
console.warn('[Summarizer] missing ARK_API_KEY or SUMMARY_MODEL, skip');
return null;
}
const transcript = recentMessages
.map((m) => `${m.role === 'user' ? '用户' : '助手'}：${(m.content || '').trim()}`)
.filter(Boolean)
.join('\n');
if (!transcript) return null;
const userContent = existingSummary
? `已有摘要：${existingSummary}\n\n新增对话：\n${transcript}`
: `对话记录：\n${transcript}`;
try {
const response = await axios.post(ARK_BASE_URL, {
model: SUMMARY_MODEL,
messages: [
{ role: 'system', content: SUMMARY_PROMPT },
{ role: 'user', content: userContent },
],
max_tokens: SUMMARY_MAX_TOKENS,
stream: false,
thinking: { type: 'enabled' },
}, {
headers: {
'Content-Type': 'application/json',
'Authorization': `Bearer ${ARK_API_KEY}`,
},
timeout: 10000,
});
const content = response.data?.choices?.[0]?.message?.content;
return content ? content.trim() : null;
} catch (err) {
console.warn('[Summarizer] LLM call failed:', err.message);
return null;
}
}
// ============ 触发检查 ============
function triggerSummarizeIfNeeded(session, sessionId) {
if (!ENABLED) return;
const turnCount = session._turnCount || 0;
if (turnCount < SUMMARIZE_EVERY_N_TURNS) return;
if (turnCount % SUMMARIZE_EVERY_N_TURNS !== 0) return;
// 异步执行，不阻塞对话
_doSummarize(session, sessionId).catch((err) => {
console.warn('[Summarizer] async summarize failed:', err.message);
});
}
async function _doSummarize(session, sessionId) {
// 获取现有摘要
const existingSummary = await redisClient.getSummary(sessionId);
// 获取最近3轮原文
let recent = await redisClient.getRecentHistory(sessionId, SUMMARIZE_EVERY_N_TURNS);
if (!recent || recent.length < 2) {
// Redis 缺失时从 DB 降级
try {
recent = await db.getHistoryForLLM(sessionId, SUMMARIZE_EVERY_N_TURNS * 2);
} catch { /* ignore */ }
}
if (!recent || recent.length < 2) return;
const summary = await summarizeConversation(existingSummary, recent);
if (!summary) return;
// 双写 Redis + MySQL
await redisClient.setSummary(sessionId, summary);
db.upsertConversationSummary(sessionId, session.userId || null, summary, {
turnCount: session._turnCount || 0,
topics: extractTopicTags(summary),
}).catch((err) => {
console.warn('[Summarizer] MySQL upsert failed:', err.message);
});
console.log(`[Summarizer] session=${sessionId} turn=${session._turnCount} summary=${summary.length}chars`);
}
// ============ 三级降级加载 ============
async function loadBestSummary(sessionId) {
// L1: Redis（~1ms）
try {
const redisSummary = await redisClient.getSummary(sessionId);
if (redisSummary) return redisSummary;
} catch { /* continue to L2 */ }
// L2: MySQL conversation_summaries（~5ms）
try {
const row = await db.getSessionSummary(sessionId);
if (row && row.summary) {
// 回填 Redis
redisClient.setSummary(sessionId, row.summary).catch(() => {});
return row.summary;
}
} catch { /* continue to L3 */ }
// L3: 降级到现有确定性摘要（由调用方处理）
return null;
}
// ============ 会话结束时持久化 ============
async function persistFinalSummary(session) {
if (!ENABLED) return;
if (!session._turnCount || session._turnCount < MIN_TURNS_TO_PERSIST) return;
const sessionId = session.sessionId;
// 优先用已有的 LLM 摘要
let summary = null;
try {
summary = await redisClient.getSummary(sessionId);
} catch { /* ignore */ }
// 如果还没生成过摘要（对话不足3轮但>=2轮），立刻生成一次
if (!summary) {
let history = await redisClient.getRecentHistory(sessionId, 5);
if (!history || history.length < 2) {
try {
history = await db.getHistoryForLLM(sessionId, 10);
} catch { /* ignore */ }
}
if (history && history.length >= 2) {
summary = await summarizeConversation(null, history);
}
}
if (!summary) return;
// 写入 MySQL（Redis 无需写，会话已结束）
await db.upsertConversationSummary(sessionId, session.userId || null, summary, {
turnCount: session._turnCount || 0,
topics: extractTopicTags(summary),
});
console.log(`[Summarizer] persisted final summary for session=${sessionId} turns=${session._turnCount}`);
}
// ============ 话题标签提取 ============
const PRODUCT_KEYWORDS = [
'活力健', '基础三合一', '肽美', '小红', '大白', '小白',
'FitLine', 'PM', '一成系统', '大沃',
'心脏宝', '关节灵', '益力康', '免疫宝', '纤体乐',
'奥适宝', 'Optimal Set', 'Basics', 'Activize', 'Beauty',
'Restorate', 'PowerCocktail', 'ProShape',
];
function extractTopicTags(text) {
if (!text) return null;
const tags = new Set();
for (const kw of PRODUCT_KEYWORDS) {
if (text.includes(kw)) {
tags.add(kw);
}
}
const arr = [...tags].slice(0, 10);
return arr.length > 0 ? arr : null;
}
module.exports = {
triggerSummarizeIfNeeded,
summarizeConversation,
loadBestSummary,
persistFinalSummary,
extractTopicTags,
SUMMARIZE_EVERY_N_TURNS,
};

93
test2/server/services/nativeVoiceGateway.js
View File

@@ -27,21 +27,27 @@ const {
const ToolExecutor = require('./toolExecutor');
const {
DEFAULT_VOICE_ASSISTANT_PROFILE,
DEFAULT_CONSULTANT_CONTACT,
resolveAssistantProfile,
getAssistantDisplayName,
buildVoiceSystemRole,
buildVoiceGreeting,
} = require('./assistantProfileConfig');
const { getAssistantProfile } = require('./assistantProfileService');
const redisClient = require('./redisClient');
const { checkProductLinkTrigger } = require('./productLinkTrigger');
const { triggerSummarizeIfNeeded, persistFinalSummary } = require('./conversationSummarizer');
const sessions = new Map();
const CONSULTANT_REFERRAL_PATTERN = /咨询(?:专业|你的)?顾问|健康管理顾问|联系顾问|一对一指导|咨询专业|咨询医生|咨询营养师|咨询专业人士|建议.*咨询|问问医生|问问.*营养师/;
const IDLE_TIMEOUT_MS = 5 * 60 * 1000;
const AUDIO_KEEPALIVE_INTERVAL_MS = 20 * 1000;
// 3200 bytes ≈ 66ms of silence at 24kHz s16le mono (larger frame to ensure S2S acceptance)
const SILENT_AUDIO_FRAME = Buffer.alloc(3200, 0);
const DEFAULT_VOICE_BOT_NAME = DEFAULT_VOICE_ASSISTANT_PROFILE.nickname;
const DEFAULT_VOICE_BOT_NAME = getAssistantDisplayName(DEFAULT_VOICE_ASSISTANT_PROFILE);
const DEFAULT_VOICE_SYSTEM_ROLE = buildVoiceSystemRole();
@@ -187,6 +193,7 @@ function persistUserSpeech(session, text) {
session.lastPersistedUserAt = now;
session.latestUserText = cleanText;
session.latestUserTurnSeq = (session.latestUserTurnSeq || 0) + 1;
session._turnCount = (session._turnCount || 0) + 1;
resetIdleTimer(session);
db.addMessage(session.sessionId, 'user', cleanText, 'voice_asr').catch((e) => console.warn('[NativeVoice][DB] add user failed:', e.message));
redisClient.pushMessage(session.sessionId, { role: 'user', content: cleanText, source: 'voice_asr' }).catch(() => {});
@@ -223,6 +230,24 @@ function persistAssistantSpeech(session, text, { source = 'voice_bot', toolName
toolName,
sequence: `native_assistant_${now}`,
});
// 异步触发摘要检查（每N轮）
if (persistToDb) {
triggerSummarizeIfNeeded(session, session.sessionId);
}
if (CONSULTANT_REFERRAL_PATTERN.test(cleanText)) {
const contact = session.consultantContact || DEFAULT_CONSULTANT_CONTACT;
if (contact.mobile || contact.wx_qr_code || contact.wechat_id) {
console.log(`[NativeVoice] consultant referral detected session=${session.sessionId} text=${JSON.stringify(cleanText.slice(0, 80))}`);
sendJson(session.client, {
type: 'consultant_contact',
name: contact.name || '大沃专业健康管理顾问',
mobile: contact.mobile || '',
wx_qr_code: contact.wx_qr_code || '',
wechat_id: contact.wechat_id || '',
message: '如需个性化健康建议，可联系大沃专业健康管理顾问',
});
}
}
return true;
}
@@ -250,6 +275,7 @@ function flushAssistantStream(session, { source = 'voice_bot', toolName = null,
return persistAssistantSpeech(session, fullText, { source, toolName, meta });
}
async function loadHandoffSummaryForVoice(session) {
try {
const history = await db.getHistoryForLLM(session.sessionId, 10);
@@ -404,6 +430,7 @@ function clearUpstreamSuppression(session) {
session.pendingAssistantSource = null;
session.pendingAssistantToolName = null;
session.pendingAssistantMeta = null;
session._pendingEvidencePack = null;
session.pendingAssistantTurnSeq = 0;
session.blockUpstreamAudio = false;
sendJson(session.client, { type: 'assistant_pending', active: false });
@@ -487,7 +514,18 @@ async function processReply(session, text, turnSeq = session.latestUserTurnSeq |
if (!resolveResult) {
resolveResult = await resolveReply(session.sessionId, session, cleanText);
}
const { delivery, speechText, ragItems, source, toolName, routeDecision, responseMeta } = resolveResult;
const { delivery, speechText, ragItems, source, toolName, routeDecision, responseMeta, evidencePack } = resolveResult;
// 产品链接触发检测：用户请求查看产品详情时推送对应链接
const productLinkResult = checkProductLinkTrigger(cleanText);
if (productLinkResult.triggered && productLinkResult.product) {
console.log(`[NativeVoice] product link triggered session=${session.sessionId} product=${productLinkResult.product.name}`);
sendJson(session.client, {
type: 'product_link',
product: productLinkResult.product.name,
link: productLinkResult.product.link,
description: productLinkResult.product.description,
});
}
if (activeTurnSeq !== (session.latestUserTurnSeq || 0)) {
console.log(`[NativeVoice] stale reply ignored session=${session.sessionId} activeTurn=${activeTurnSeq} latestTurn=${session.latestUserTurnSeq || 0}`);
clearUpstreamSuppression(session);
@@ -501,6 +539,7 @@ async function processReply(session, text, turnSeq = session.latestUserTurnSeq |
}
session.blockUpstreamAudio = false;
session._lastPartialAt = 0;
session._pendingEvidencePack = null;
session.awaitingUpstreamReply = true;
session.pendingAssistantSource = 'voice_bot';
session.pendingAssistantToolName = null;
@@ -515,7 +554,7 @@ async function processReply(session, text, turnSeq = session.latestUserTurnSeq |
}
session.discardNextAssistantResponse = true;
sendJson(session.client, { type: 'tts_reset', reason: 'knowledge_hit' });
const ragContent = (ragItems || []).filter((item) => item && item.content);
const ragContent = (ragItems || []).filter((item) => item && item.content && item.kind !== 'context');
if (ragContent.length > 0) {
console.log(`[NativeVoice] processReply sending external_rag to S2S session=${session.sessionId} route=${routeDecision?.route || 'unknown'} items=${ragContent.length}`);
// KB话题记忆：记录本轮用户原始问题和时间戳，用于保护窗口和追问enrichment
@@ -523,6 +562,7 @@ async function processReply(session, text, turnSeq = session.latestUserTurnSeq |
session._lastKbTopic = cleanText;
session._lastKbHitAt = Date.now();
}
session._pendingEvidencePack = evidencePack || null;
// 不提前发KB原文作字幕：等S2S event 351返回实际语音文本后再更新字幕
// 这样字幕和语音保持一致（S2S会基于RAG内容生成自然口语化的回答）
session._pendingExternalRagReply = true;
@@ -708,6 +748,7 @@ function handleUpstreamMessage(session, data) {
session.pendingAssistantSource = null;
session.pendingAssistantToolName = null;
session.pendingAssistantMeta = null;
session._pendingEvidencePack = null;
console.log(`[NativeVoice] duplicate assistant final ignored (351) session=${session.sessionId} turn=${pendingAssistantTurnSeq}`);
return;
}
@@ -725,30 +766,17 @@ function handleUpstreamMessage(session, data) {
if (session._pendingExternalRagReply) {
session._pendingExternalRagReply = false;
}
// 品牌安全检测：最终助手文本包含有害内容时，阻断音频并注入安全回复
if (isBrandHarmful(assistantText)) {
console.warn(`[NativeVoice][SafeGuard] harmful content in final assistant text, blocking session=${session.sessionId} text=${JSON.stringify(assistantText.slice(0, 120))}`);
console.log(`[NativeVoice] upstream assistant session=${session.sessionId} text=${JSON.stringify(assistantText.slice(0, 120))}`);
session.lastDeliveredAssistantTurnSeq = pendingAssistantTurnSeq;
persistAssistantSpeech(session, assistantText, {
source: pendingAssistantSource,
toolName: pendingAssistantToolName,
meta: pendingAssistantMeta,
});
// KB回复完成后重新阻断音频，防止下一个问题的S2S默认回复在early block前泄露
if (session.currentTtsType === 'external_rag') {
session.blockUpstreamAudio = true;
sendJson(session.client, { type: 'tts_reset', reason: 'harmful_blocked' });
const safeReply = getVoiceSafeReply();
session.lastDeliveredAssistantTurnSeq = pendingAssistantTurnSeq;
persistAssistantSpeech(session, safeReply, { source: 'voice_bot' });
sendSpeechText(session, safeReply).catch((err) => {
console.warn('[NativeVoice][SafeGuard] sendSpeechText failed:', err.message);
});
} else {
console.log(`[NativeVoice] upstream assistant session=${session.sessionId} text=${JSON.stringify(assistantText.slice(0, 120))}`);
session.lastDeliveredAssistantTurnSeq = pendingAssistantTurnSeq;
persistAssistantSpeech(session, assistantText, {
source: pendingAssistantSource,
toolName: pendingAssistantToolName,
meta: pendingAssistantMeta,
});
// KB回复完成后重新阻断音频，防止下一个问题的S2S默认回复在early block前泄露
if (session.currentTtsType === 'external_rag') {
session.blockUpstreamAudio = true;
console.log(`[NativeVoice] re-blocked after KB response session=${session.sessionId}`);
}
console.log(`[NativeVoice] re-blocked after KB response session=${session.sessionId}`);
}
} else {
const didFlush = flushAssistantStream(session, {
@@ -763,6 +791,7 @@ function handleUpstreamMessage(session, data) {
session.pendingAssistantSource = null;
session.pendingAssistantToolName = null;
session.pendingAssistantMeta = null;
session._pendingEvidencePack = null;
return;
}
@@ -845,6 +874,7 @@ function handleUpstreamMessage(session, data) {
session.pendingAssistantSource = null;
session.pendingAssistantToolName = null;
session.pendingAssistantMeta = null;
session._pendingEvidencePack = null;
console.log(`[NativeVoice] duplicate assistant final ignored (559) session=${session.sessionId} turn=${pendingAssistantTurnSeq}`);
return;
}
@@ -862,6 +892,7 @@ function handleUpstreamMessage(session, data) {
session.pendingAssistantSource = null;
session.pendingAssistantToolName = null;
session.pendingAssistantMeta = null;
session._pendingEvidencePack = null;
return;
}
@@ -1042,7 +1073,7 @@ function attachClientHandlers(session) {
...((parsed.assistantProfile && typeof parsed.assistantProfile === 'object') ? parsed.assistantProfile : {}),
});
session.assistantProfile = assistantProfile;
session.botName = parsed.botName || assistantProfile.nickname || DEFAULT_VOICE_BOT_NAME;
session.botName = parsed.botName || getAssistantDisplayName(assistantProfile) || DEFAULT_VOICE_BOT_NAME;
session.systemRole = buildVoiceSystemRole(assistantProfile);
session.speakingStyle = parsed.speakingStyle || session.speakingStyle || DEFAULT_VOICE_SPEAKING_STYLE;
session.speaker = parsed.speaker || process.env.VOLC_S2S_SPEAKER_ID || 'zh_female_vv_jupiter_bigtts';
@@ -1089,6 +1120,10 @@ function attachClientHandlers(session) {
if (session.upstream && session.upstream.readyState === WebSocket.OPEN) {
session.upstream.close();
}
// 会话结束时持久化摘要到 MySQL
persistFinalSummary(session).catch((err) => {
console.warn('[NativeVoice] persistFinalSummary failed:', err.message);
});
sessions.delete(session.sessionId);
});
}
@@ -1164,6 +1199,7 @@ function createSession(client, sessionId) {
_echoLogOnce: false,
_fillerActive: false,
_pendingExternalRagReply: false,
_pendingEvidencePack: null,
_lastPartialAt: 0,
pendingKbPrequery: null,
_kbPrequeryText: '',
@@ -1171,7 +1207,7 @@ function createSession(client, sessionId) {
_lastKbTopic: '',
_lastKbHitAt: 0,
assistantProfile,
botName: assistantProfile.nickname,
botName: getAssistantDisplayName(assistantProfile),
systemRole: buildVoiceSystemRole(assistantProfile),
speakingStyle: DEFAULT_VOICE_SPEAKING_STYLE,
speaker: process.env.VOLC_S2S_SPEAKER_ID || 'zh_female_vv_jupiter_bigtts',
@@ -1201,6 +1237,7 @@ function createSession(client, sessionId) {
_lastFinalNormalized: '',
_lastFinalAt: 0,
_audioKeepaliveTimer: null,
_turnCount: 0,
};
sessions.set(sessionId, session);
attachClientHandlers(session);

74
test2/server/services/realtimeDialogRouting.js
View File

@@ -1,7 +1,9 @@
const ToolExecutor = require('./toolExecutor');
const db = require('../db');
const redisClient = require('./redisClient');
const knowledgeQueryResolver = require('./knowledgeQueryResolver');
const { hasKnowledgeRouteKeyword } = require('./knowledgeKeywords');
const { loadBestSummary } = require('./conversationSummarizer');
function normalizeTextForSpeech(text) {
return (text || '')
@@ -64,7 +66,8 @@ function estimateSpeechDurationMs(text) {
}
function normalizeKnowledgeAlias(text) {
return String(text || '')
const semanticNormalized = knowledgeQueryResolver.normalizeKnowledgeText(text);
return String(semanticNormalized || '')
.replace(/一成[，,、。！？\s]+系统/g, '一成系统')
.replace(/X{2}系统/gi, '一成系统')
.replace(/[\u4e00-\u9fff]{1,3}(?:成|城|程|诚|乘|声|生)[，,、\s]*系统/g, '一成系统')
@@ -77,7 +80,7 @@ function normalizeKnowledgeAlias(text) {
function hasKnowledgeKeyword(text) {
const normalized = normalizeKnowledgeAlias(text).replace(/\s+/g, '');
return hasKnowledgeRouteKeyword(normalized);
return hasKnowledgeRouteKeyword(normalized) || knowledgeQueryResolver.hasExplicitKnowledgeEntity(normalized, { skipAsrCorrection: true });
}
function isKnowledgeFollowUp(text) {
@@ -253,7 +256,7 @@ function extractToolResultText(toolName, toolResult) {
return '知识库已配置但方舟LLM端点未就绪，暂时无法检索，请稍后再试。';
}
if (toolResult.results && Array.isArray(toolResult.results)) {
return toolResult.results.map((item) => item.content || JSON.stringify(item)).join('\n');
return toolResult.results.filter((item) => item.kind !== 'instruction' && item.kind !== 'context').map((item) => item.content || JSON.stringify(item)).join('\n');
}
if (typeof toolResult === 'string') return toolResult;
if (toolResult.error) return toolResult.error;
@@ -272,24 +275,23 @@ async function resolveReply(sessionId, session, text) {
// 快速路径：知识库候选先尝试无context的热答案/缓存命中，跳过DB查询（省50-200ms）
if (shouldForceKnowledgeRoute(originalText)) {
const fastResult = await ToolExecutor.execute('search_knowledge', { query: originalText, response_mode: 'answer', session_id: sessionId, _session: session }, []);
const fastResult = await ToolExecutor.execute('search_knowledge', { query: originalText, session_id: sessionId, _session: session }, []);
if (fastResult && fastResult.hit) {
const replyText = extractToolResultText('search_knowledge', fastResult);
const ragItems = fastResult.hit && Array.isArray(fastResult.results)
? fastResult.results.filter(i => i && i.content).map(i => ({ title: i.title || '知识库结果', content: i.content }))
const evidenceItems = Array.isArray(fastResult?.evidence_pack?.items) && fastResult.evidence_pack.items.length > 0
? fastResult.evidence_pack.items
: (Array.isArray(fastResult.results) ? fastResult.results : []);
const ragItems = fastResult.hit && Array.isArray(evidenceItems)
? evidenceItems.filter(i => i && i.content && i.kind !== 'context').map(i => ({ ...i }))
: [];
console.log(`[resolveReply] fast-path hit in ${Date.now() - _resolveStart}ms session=${sessionId} source=${fastResult.cache_hit ? 'cache' : 'direct'} mode=${fastResult.retrieval_mode || 'answer'}`);
console.log(`[resolveReply] fast-path hit in ${Date.now() - _resolveStart}ms session=${sessionId} source=${fastResult.cache_hit ? 'cache' : 'direct'} mode=raw`);
if (ragItems.length > 0) {
session.handoffSummaryUsed = true;
// raw 模式：ragItems 已包含上下文 + 多个 KB 片段，直接透传
const isRawMode = fastResult.retrieval_mode === 'raw';
const finalRagItems = isRawMode
? ragItems
: [{ title: '知识库结果', content: normalizeTextForSpeech(replyText).replace(/^(根据知识库信息[，,：:\s]*|根据.*?[，,]\s*)/i, '') || replyText }];
return {
delivery: 'external_rag',
speechText: '',
ragItems: finalRagItems,
ragItems,
evidencePack: fastResult.evidence_pack || null,
source: 'voice_tool',
toolName: 'search_knowledge',
routeDecision: { route: 'search_knowledge', args: { query: originalText } },
@@ -311,7 +313,7 @@ async function resolveReply(sessionId, session, text) {
const _dbStart = Date.now();
let recentMessages = null;
if (process.env.ENABLE_REDIS_CONTEXT !== 'false') {
const redisHistory = await redisClient.getRecentHistory(sessionId, 5).catch(() => null);
const redisHistory = await redisClient.getRecentHistory(sessionId, 3).catch(() => null);
if (redisHistory && redisHistory.length > 0) {
recentMessages = redisHistory;
const _dbMs = Date.now() - _dbStart;
@@ -319,7 +321,7 @@ async function resolveReply(sessionId, session, text) {
}
}
if (!recentMessages) {
recentMessages = await db.getRecentMessages(sessionId, 10).catch(() => []);
recentMessages = await db.getRecentMessages(sessionId, 6).catch(() => []);
const _dbMs = Date.now() - _dbStart;
if (_dbMs > 50) console.log(`[resolveReply] DB getRecentMessages took ${_dbMs}ms session=${sessionId}`);
}
@@ -329,7 +331,12 @@ async function resolveReply(sessionId, session, text) {
const baseContext = scopedMessages
.filter((item) => item && (item.role === 'user' || item.role === 'assistant'))
.map((item) => ({ role: item.role, content: item.content }));
const context = withHandoffSummary(session, baseContext);
// 注入对话摘要（三级降级：Redis → MySQL → null）
const summary = await loadBestSummary(sessionId).catch(() => null);
const contextWithSummary = summary
? [{ role: 'system', content: `[历史对话摘要] ${summary}` }, ...baseContext]
: baseContext;
const context = withHandoffSummary(session, contextWithSummary);
let routeDecision = getRuleBasedDirectRouteDecision(originalText);
// KB-First: 所有非闲聊查询强制先走知识库，KB不命中再交给S2S自由回答
if (routeDecision.route === 'chat' && !isPureChitchat(originalText)) {
@@ -358,7 +365,7 @@ async function resolveReply(sessionId, session, text) {
toolName = routeDecision.route;
source = 'voice_tool';
const toolArgs = toolName === 'search_knowledge'
? { ...(routeDecision.args || {}), response_mode: 'answer', session_id: sessionId, _session: session }
? { ...(routeDecision.args || {}), session_id: sessionId, _session: session }
: routeDecision.args;
const metaToolArgs = toolArgs && typeof toolArgs === 'object'
? Object.fromEntries(Object.entries(toolArgs).filter(([key]) => key !== '_session'))
@@ -380,14 +387,15 @@ async function resolveReply(sessionId, session, text) {
latency_ms: toolResult?.latency_ms || null,
};
const evidenceItems = Array.isArray(toolResult?.evidence_pack?.items) && toolResult.evidence_pack.items.length > 0
? toolResult.evidence_pack.items
: (Array.isArray(toolResult?.results) ? toolResult.results : []);
const ragItems = toolName === 'search_knowledge'
? (toolResult?.hit && Array.isArray(toolResult?.results)
? toolResult.results
.filter((item) => item && item.content)
.map((item) => ({
title: item.title || '知识库结果',
content: item.content,
}))
? (toolResult?.hit && Array.isArray(evidenceItems)
? evidenceItems
.filter((item) => item && item.content && item.kind !== 'context')
.map((item) => ({ ...item }))
: [])
: (!toolResult?.error && replyText
? [{ title: `${toolName}结果`, content: replyText }]
@@ -395,23 +403,11 @@ async function resolveReply(sessionId, session, text) {
if (ragItems.length > 0) {
session.handoffSummaryUsed = true;
const isRawMode = toolResult?.retrieval_mode === 'raw';
let finalRagItems = ragItems;
if (toolName === 'search_knowledge' && !isRawMode) {
// 旧模式：LLM 加工过的文本，清理后合并为单条
const speechText = normalizeTextForSpeech(replyText);
if (speechText) {
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 }];
}
}
// raw 模式：ragItems 已包含上下文 + 多个 KB 片段，直接透传给 S2S
return {
delivery: 'external_rag',
speechText: '',
ragItems: finalRagItems,
ragItems,
evidencePack: toolResult?.evidence_pack || null,
source,
toolName,
routeDecision,
@@ -428,6 +424,7 @@ async function resolveReply(sessionId, session, text) {
delivery: 'external_rag',
speechText: '',
ragItems: [{ title: '品牌保护', content: safeReply }],
evidencePack: toolResult?.evidence_pack || null,
source: 'voice_tool',
toolName: 'search_knowledge',
routeDecision,
@@ -443,6 +440,7 @@ async function resolveReply(sessionId, session, text) {
delivery: 'external_rag',
speechText: '',
ragItems: [{ title: '知识库未命中', content: honestReply }],
evidencePack: toolResult?.evidence_pack || null,
source: 'voice_tool',
toolName: 'search_knowledge',
routeDecision,

29
test2/server/services/redisClient.js
View File

@@ -10,6 +10,7 @@ const HISTORY_MAX_LEN = 10; // 5轮 × 2条/轮
const HISTORY_TTL_S = 1800; // 30分钟
const KB_CACHE_HIT_TTL_S = 300; // 5分钟
const KB_CACHE_NOHIT_TTL_S = 120; // 2分钟
const SUMMARY_TTL_S = parseInt(process.env.SUMMARY_REDIS_TTL_SECONDS) || 7200; // 2小时
// ============ 连接管理 ============
let client = null;
@@ -128,6 +129,32 @@ async function clearSession(sessionId) {
}
}
// ============ 对话摘要 ============
const summaryKey = (sessionId) => `voice:summary:${sessionId}`;
async function setSummary(sessionId, summary) {
if (!isAvailable() || !summary) return false;
try {
const key = summaryKey(sessionId);
await client.set(key, summary, 'EX', SUMMARY_TTL_S);
return true;
} catch (err) {
console.warn('[Redis] setSummary failed:', err.message);
return false;
}
}
async function getSummary(sessionId) {
if (!isAvailable()) return null;
try {
const key = summaryKey(sessionId);
return await client.get(key);
} catch (err) {
console.warn('[Redis] getSummary failed:', err.message);
return null;
}
}
// ============ KB 缓存 ============
const kbCacheKey = (cacheKey) => `kb_cache:${cacheKey}`;
@@ -178,6 +205,8 @@ module.exports = {
pushMessage,
getRecentHistory,
clearSession,
setSummary,
getSummary,
setKbCache,
getKbCache,
disconnect,