mirror of
https://github.com/samanhappy/mcphub.git
synced 2026-01-03 05:10:09 -05:00
refine docs
This commit is contained in:
samanhappy
@@ -1,691 +1,367 @@
|
||||
---
|
||||
title: '智能路由'
|
||||
description: '自动负载均衡和请求路由到最佳的 MCP 服务器实例'
|
||||
description: '使用向量语义搜索的 AI 工具发现系统'
|
||||
---
|
||||
|
||||
## 概述
|
||||
|
||||
MCPHub 的智能路由系统自动将传入请求路由到最适合的 MCP 服务器实例。系统考虑服务器负载、响应时间、功能可用性和业务规则来做出路由决策。
|
||||
智能路由是 MCPHub 的智能工具发现系统，它使用向量语义搜索来自动找到与任何给定任务最相关的工具。AI 客户端无需手动指定使用哪些工具，只需描述他们想要完成的任务，智能路由就会识别并提供对最合适工具的访问。
|
||||
|
||||
## 路由策略
|
||||
## 智能路由的工作原理
|
||||
|
||||
### 轮询路由
|
||||
### 1. 工具索引
|
||||
|
||||
最简单的路由策略，按顺序分发请求：
|
||||
当服务器启动时，智能路由会自动：
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "round-robin",
|
||||
"targets": [
|
||||
{
|
||||
"serverId": "server-1",
|
||||
"weight": 1,
|
||||
"enabled": true
|
||||
},
|
||||
{
|
||||
"serverId": "server-2",
|
||||
"weight": 1,
|
||||
"enabled": true
|
||||
},
|
||||
{
|
||||
"serverId": "server-3",
|
||||
"weight": 1,
|
||||
"enabled": true
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
- 从 MCP 服务器发现所有可用工具
|
||||
- 提取工具元数据（名称、描述、参数）
|
||||
- 将工具信息转换为向量嵌入
|
||||
- 使用 pgvector 将嵌入存储在 PostgreSQL 中
|
||||
|
||||
### 加权轮询
|
||||
### 2. 语义搜索
|
||||
|
||||
基于服务器容量分配不同权重：
|
||||
当进行查询时：
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "weighted-round-robin",
|
||||
"targets": [
|
||||
{
|
||||
"serverId": "high-performance-server",
|
||||
"weight": 3,
|
||||
"specs": {
|
||||
"cpu": "8 cores",
|
||||
"memory": "32GB"
|
||||
}
|
||||
},
|
||||
{
|
||||
"serverId": "standard-server-1",
|
||||
"weight": 2,
|
||||
"specs": {
|
||||
"cpu": "4 cores",
|
||||
"memory": "16GB"
|
||||
}
|
||||
},
|
||||
{
|
||||
"serverId": "standard-server-2",
|
||||
"weight": 1,
|
||||
"specs": {
|
||||
"cpu": "2 cores",
|
||||
"memory": "8GB"
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
- 用户查询被转换为向量嵌入
|
||||
- 相似性搜索使用余弦相似度找到匹配的工具
|
||||
- 动态阈值过滤掉不相关的结果
|
||||
- 结果按相关性得分排序
|
||||
|
||||
### 最少连接数
|
||||
### 3. 智能过滤
|
||||
|
||||
将请求路由到当前连接数最少的服务器：
|
||||
智能路由应用多个过滤器：
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "least-connections",
|
||||
"balancingMode": "dynamic",
|
||||
"healthCheck": {
|
||||
"enabled": true,
|
||||
"interval": 10000
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
- **相关性阈值**：只返回高于相似性阈值的工具
|
||||
- **上下文感知**：考虑对话上下文
|
||||
- **工具可用性**：确保工具当前可访问
|
||||
- **权限过滤**：尊重用户访问权限
|
||||
|
||||
### 基于响应时间
|
||||
### 4. 工具执行
|
||||
|
||||
路由到响应时间最短的服务器：
|
||||
找到的工具可以直接执行：
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "fastest-response",
|
||||
"metrics": {
|
||||
"measurementWindow": "5m",
|
||||
"sampleSize": 100,
|
||||
"excludeSlowRequests": true,
|
||||
"slowRequestThreshold": "5s"
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
- 参数验证确保正确的工具使用
|
||||
- 错误处理提供有用的反馈
|
||||
- 响应格式保持一致性
|
||||
- 日志记录跟踪工具使用情况进行分析
|
||||
|
||||
## 基于功能的路由
|
||||
## 前置条件
|
||||
|
||||
### 工具特定路由
|
||||
智能路由需要比基础 MCPHub 使用更多的设置：
|
||||
|
||||
根据请求的工具类型路由到专门的服务器：
|
||||
### 必需组件
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "capability-based",
|
||||
"rules": [
|
||||
{
|
||||
"condition": {
|
||||
"tool": "filesystem"
|
||||
},
|
||||
"targets": ["filesystem-server-1", "filesystem-server-2"],
|
||||
"strategy": "least-connections"
|
||||
},
|
||||
{
|
||||
"condition": {
|
||||
"tool": "web-search"
|
||||
},
|
||||
"targets": ["search-server-1", "search-server-2"],
|
||||
"strategy": "round-robin"
|
||||
},
|
||||
{
|
||||
"condition": {
|
||||
"tool": "database"
|
||||
},
|
||||
"targets": ["db-server"],
|
||||
"strategy": "single"
|
||||
}
|
||||
],
|
||||
"fallback": {
|
||||
"targets": ["general-server-1", "general-server-2"],
|
||||
"strategy": "round-robin"
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
1. **带有 pgvector 的 PostgreSQL**：用于嵌入存储的向量数据库
|
||||
2. **嵌入服务**：OpenAI API 或兼容服务
|
||||
3. **环境配置**：正确的配置变量
|
||||
|
||||
### 内容感知路由
|
||||
## 使用智能路由
|
||||
|
||||
基于请求内容进行智能路由：
|
||||
### 智能路由端点
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "content-aware",
|
||||
"rules": [
|
||||
{
|
||||
"condition": {
|
||||
"content.language": "python"
|
||||
},
|
||||
"targets": ["python-specialized-server"],
|
||||
"reason": "Python代码分析专用服务器"
|
||||
},
|
||||
{
|
||||
"condition": {
|
||||
"content.size": "> 1MB"
|
||||
},
|
||||
"targets": ["high-memory-server"],
|
||||
"reason": "大文件处理专用服务器"
|
||||
},
|
||||
{
|
||||
"condition": {
|
||||
"content.type": "image"
|
||||
},
|
||||
"targets": ["image-processing-server"],
|
||||
"reason": "图像处理专用服务器"
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
通过特殊的 `$smart` 端点访问智能路由：
|
||||
|
||||
## 地理位置路由
|
||||
<Tabs>
|
||||
<Tab title="HTTP MCP">
|
||||
```
|
||||
http://localhost:3000/mcp/$smart
|
||||
```
|
||||
</Tab>
|
||||
|
||||
### 基于客户端位置
|
||||
<Tab title="SSE (Legacy)">
|
||||
```
|
||||
http://localhost:3000/sse/$smart
|
||||
```
|
||||
</Tab>
|
||||
</Tabs>
|
||||
|
||||
根据客户端地理位置路由到最近的服务器：
|
||||
{/* ## 性能优化
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "geo-location",
|
||||
"regions": [
|
||||
{
|
||||
"name": "北美",
|
||||
"countries": ["US", "CA", "MX"],
|
||||
"servers": ["us-east-1", "us-west-1", "ca-central-1"],
|
||||
"strategy": "least-latency"
|
||||
},
|
||||
{
|
||||
"name": "欧洲",
|
||||
"countries": ["DE", "FR", "UK", "NL"],
|
||||
"servers": ["eu-west-1", "eu-central-1"],
|
||||
"strategy": "round-robin"
|
||||
},
|
||||
{
|
||||
"name": "亚太",
|
||||
"countries": ["CN", "JP", "KR", "SG"],
|
||||
"servers": ["ap-southeast-1", "ap-northeast-1"],
|
||||
"strategy": "fastest-response"
|
||||
}
|
||||
],
|
||||
"fallback": {
|
||||
"servers": ["global-server-1"],
|
||||
"strategy": "single"
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
### 嵌入缓存
|
||||
|
||||
### 延迟优化
|
||||
智能路由缓存嵌入以提高性能：
|
||||
|
||||
```bash
|
||||
# 配置延迟监控
|
||||
curl -X PUT http://localhost:3000/api/routing/latency-config \
|
||||
-H "Content-Type: application/json" \
|
||||
-H "Authorization: Bearer $TOKEN" \
|
||||
-d '{
|
||||
# 配置缓存设置
|
||||
EMBEDDING_CACHE_TTL=3600 # 缓存 1 小时
|
||||
EMBEDDING_CACHE_SIZE=10000 # 最多缓存 10k 个嵌入
|
||||
EMBEDDING_CACHE_CLEANUP=300 # 每 5 分钟清理一次
|
||||
```
|
||||
|
||||
### 批处理
|
||||
|
||||
工具批量索引以提高效率：
|
||||
|
||||
```bash
|
||||
# 嵌入生成的批大小
|
||||
EMBEDDING_BATCH_SIZE=100
|
||||
|
||||
# 并发嵌入请求
|
||||
EMBEDDING_CONCURRENCY=5
|
||||
|
||||
# 索引更新频率
|
||||
INDEX_UPDATE_INTERVAL=3600 # 每小时重新索引
|
||||
```
|
||||
|
||||
### 数据库优化
|
||||
|
||||
为向量操作优化 PostgreSQL：
|
||||
|
||||
```sql
|
||||
-- 创建索引以获得更好的性能
|
||||
CREATE INDEX ON tool_embeddings USING hnsw (embedding vector_cosine_ops);
|
||||
|
||||
-- 调整 PostgreSQL 设置
|
||||
ALTER SYSTEM SET shared_preload_libraries = 'vector';
|
||||
ALTER SYSTEM SET max_connections = 200;
|
||||
ALTER SYSTEM SET shared_buffers = '256MB';
|
||||
ALTER SYSTEM SET effective_cache_size = '1GB';
|
||||
```
|
||||
|
||||
## 监控和分析
|
||||
|
||||
### 智能路由指标
|
||||
|
||||
监控智能路由性能：
|
||||
|
||||
```bash
|
||||
# 获取智能路由统计信息
|
||||
curl http://localhost:3000/api/smart-routing/stats \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN"
|
||||
```
|
||||
|
||||
响应包括：
|
||||
|
||||
- 查询计数和频率
|
||||
- 平均响应时间
|
||||
- 嵌入缓存命中率
|
||||
- 最受欢迎的工具
|
||||
- 查询模式
|
||||
|
||||
### 工具使用分析
|
||||
|
||||
跟踪哪些工具被发现和使用：
|
||||
|
||||
```bash
|
||||
# 获取工具使用分析
|
||||
curl http://localhost:3000/api/smart-routing/analytics \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN"
|
||||
```
|
||||
|
||||
指标包括：
|
||||
|
||||
- 工具发现率
|
||||
- 执行成功率
|
||||
- 用户满意度评分
|
||||
- 查询到执行的转换率
|
||||
|
||||
### 性能监控
|
||||
|
||||
监控系统性能：
|
||||
|
||||
```bash
|
||||
# 数据库性能
|
||||
curl http://localhost:3000/api/smart-routing/db-stats \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN"
|
||||
|
||||
# 嵌入服务状态
|
||||
curl http://localhost:3000/api/smart-routing/embedding-stats \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN"
|
||||
```
|
||||
|
||||
## 高级功能
|
||||
|
||||
### 自定义嵌入
|
||||
|
||||
使用自定义嵌入模型：
|
||||
|
||||
```bash
|
||||
# Hugging Face 模型
|
||||
EMBEDDING_SERVICE=huggingface
|
||||
HUGGINGFACE_MODEL=sentence-transformers/all-MiniLM-L6-v2
|
||||
HUGGINGFACE_API_KEY=your_api_key
|
||||
|
||||
# 本地嵌入服务
|
||||
EMBEDDING_SERVICE=local
|
||||
EMBEDDING_SERVICE_URL=http://localhost:8080/embeddings
|
||||
```
|
||||
|
||||
### 查询增强
|
||||
|
||||
增强查询以获得更好的结果：
|
||||
|
||||
```json
|
||||
{
|
||||
"queryEnhancement": {
|
||||
"enabled": true,
|
||||
"measurementInterval": 30000,
|
||||
"regions": [
|
||||
{"id": "us-east", "endpoint": "ping.us-east.example.com"},
|
||||
{"id": "eu-west", "endpoint": "ping.eu-west.example.com"},
|
||||
{"id": "ap-southeast", "endpoint": "ping.ap-southeast.example.com"}
|
||||
],
|
||||
"routing": {
|
||||
"preferLowLatency": true,
|
||||
"maxLatencyThreshold": "200ms",
|
||||
"fallbackOnTimeout": true
|
||||
}
|
||||
}'
|
||||
```
|
||||
|
||||
## 负载感知路由
|
||||
|
||||
### 实时负载监控
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "load-aware",
|
||||
"loadMetrics": {
|
||||
"cpu": {
|
||||
"threshold": 80,
|
||||
"weight": 0.4
|
||||
},
|
||||
"memory": {
|
||||
"threshold": 85,
|
||||
"weight": 0.3
|
||||
},
|
||||
"connections": {
|
||||
"threshold": 1000,
|
||||
"weight": 0.2
|
||||
},
|
||||
"responseTime": {
|
||||
"threshold": "2s",
|
||||
"weight": 0.1
|
||||
}
|
||||
},
|
||||
"adaptation": {
|
||||
"enabled": true,
|
||||
"adjustmentInterval": 60000,
|
||||
"emergencyThreshold": 95
|
||||
}
|
||||
"expandAcronyms": true,
|
||||
"addSynonyms": true,
|
||||
"contextualExpansion": true
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### 预测性负载均衡
|
||||
### 结果过滤
|
||||
|
||||
基于条件过滤结果：
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "predictive",
|
||||
"prediction": {
|
||||
"algorithm": "linear-regression",
|
||||
"trainingWindow": "7d",
|
||||
"predictionHorizon": "1h",
|
||||
"factors": ["historical_load", "time_of_day", "day_of_week", "seasonal_patterns"]
|
||||
},
|
||||
"adaptation": {
|
||||
"preemptiveScaling": true,
|
||||
"scaleUpThreshold": 70,
|
||||
"scaleDownThreshold": 30
|
||||
}
|
||||
"resultFiltering": {
|
||||
"minRelevanceScore": 0.7,
|
||||
"maxResults": 10,
|
||||
"preferredServers": ["fetch", "playwright"],
|
||||
"excludeServers": ["deprecated-server"]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## 故障转移和恢复
|
||||
### 反馈学习
|
||||
|
||||
### 自动故障转移
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "high-availability",
|
||||
"failover": {
|
||||
"enabled": true,
|
||||
"detection": {
|
||||
"healthCheckFailures": 3,
|
||||
"timeoutThreshold": "10s",
|
||||
"checkInterval": 5000
|
||||
},
|
||||
"recovery": {
|
||||
"automaticRecovery": true,
|
||||
"recoveryChecks": 5,
|
||||
"recoveryInterval": 30000
|
||||
}
|
||||
},
|
||||
"clusters": [
|
||||
{
|
||||
"name": "primary",
|
||||
"servers": ["server-1", "server-2"],
|
||||
"priority": 1
|
||||
},
|
||||
{
|
||||
"name": "secondary",
|
||||
"servers": ["backup-server-1", "backup-server-2"],
|
||||
"priority": 2
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### 断路器模式
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"circuitBreaker": {
|
||||
"enabled": true,
|
||||
"failureThreshold": 10,
|
||||
"timeWindow": 60000,
|
||||
"halfOpenRetries": 3,
|
||||
"fallback": {
|
||||
"type": "cached-response",
|
||||
"ttl": 300000
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## 会话亲和性
|
||||
|
||||
### 粘性会话
|
||||
|
||||
保持用户会话与特定服务器的关联：
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "session-affinity",
|
||||
"affinity": {
|
||||
"type": "cookie",
|
||||
"cookieName": "mcphub-server-id",
|
||||
"ttl": 3600000,
|
||||
"fallbackOnUnavailable": true
|
||||
},
|
||||
"sessionStore": {
|
||||
"type": "redis",
|
||||
"config": {
|
||||
"host": "localhost",
|
||||
"port": 6379,
|
||||
"db": 1
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### 基于用户 ID 的路由
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "user-based",
|
||||
"userRouting": {
|
||||
"algorithm": "consistent-hashing",
|
||||
"hashFunction": "sha256",
|
||||
"virtualNodes": 100,
|
||||
"replicationFactor": 2
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## 动态路由配置
|
||||
|
||||
### 运行时配置更新
|
||||
基于用户反馈改进结果：
|
||||
|
||||
```bash
|
||||
# 更新路由配置
|
||||
curl -X PUT http://localhost:3000/api/routing/config \
|
||||
# 对搜索结果提供反馈
|
||||
curl -X POST http://localhost:3000/api/smart-routing/feedback \
|
||||
-H "Content-Type: application/json" \
|
||||
-H "Authorization: Bearer $TOKEN" \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN" \
|
||||
-d '{
|
||||
"strategy": "weighted-round-robin",
|
||||
"weights": {
|
||||
"server-1": 3,
|
||||
"server-2": 2,
|
||||
"server-3": 1
|
||||
},
|
||||
"applyImmediately": true
|
||||
"queryId": "search-123",
|
||||
"toolName": "fetch_html",
|
||||
"rating": 5,
|
||||
"successful": true,
|
||||
"comments": "完美适合这个任务的工具"
|
||||
}'
|
||||
```
|
||||
|
||||
### A/B 测试路由
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "ab-testing",
|
||||
"experiments": [
|
||||
{
|
||||
"name": "new-algorithm-test",
|
||||
"enabled": true,
|
||||
"trafficSplit": {
|
||||
"control": 70,
|
||||
"variant": 30
|
||||
},
|
||||
"rules": {
|
||||
"control": {
|
||||
"strategy": "round-robin",
|
||||
"servers": ["stable-server-1", "stable-server-2"]
|
||||
},
|
||||
"variant": {
|
||||
"strategy": "ai-optimized",
|
||||
"servers": ["experimental-server-1"]
|
||||
}
|
||||
},
|
||||
"metrics": ["response_time", "error_rate", "user_satisfaction"]
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## 路由分析和监控
|
||||
|
||||
### 实时路由指标
|
||||
|
||||
```bash
|
||||
# 获取路由统计
|
||||
curl -X GET http://localhost:3000/api/routing/metrics \
|
||||
-H "Authorization: Bearer $TOKEN"
|
||||
```
|
||||
|
||||
响应示例：
|
||||
|
||||
```json
|
||||
{
|
||||
"timestamp": "2024-01-01T12:00:00Z",
|
||||
"totalRequests": 15420,
|
||||
"routingDistribution": {
|
||||
"server-1": { "requests": 6168, "percentage": 40 },
|
||||
"server-2": { "requests": 4626, "percentage": 30 },
|
||||
"server-3": { "requests": 3084, "percentage": 20 },
|
||||
"backup-server": { "requests": 1542, "percentage": 10 }
|
||||
},
|
||||
"performance": {
|
||||
"avgResponseTime": "245ms",
|
||||
"p95ResponseTime": "580ms",
|
||||
"errorRate": "0.3%"
|
||||
},
|
||||
"failovers": {
|
||||
"total": 2,
|
||||
"byServer": {
|
||||
"server-2": 1,
|
||||
"server-3": 1
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### 路由决策日志
|
||||
|
||||
```bash
|
||||
# 启用路由决策日志
|
||||
curl -X PUT http://localhost:3000/api/routing/logging \
|
||||
-H "Content-Type: application/json" \
|
||||
-H "Authorization: Bearer $TOKEN" \
|
||||
-d '{
|
||||
"enabled": true,
|
||||
"level": "info",
|
||||
"includeDecisionFactors": true,
|
||||
"sampleRate": 0.1
|
||||
}'
|
||||
```
|
||||
|
||||
## 自定义路由规则
|
||||
|
||||
### 基于业务逻辑的路由
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "custom-rules",
|
||||
"rules": [
|
||||
{
|
||||
"name": "premium-users",
|
||||
"priority": 1,
|
||||
"condition": "user.tier === 'premium'",
|
||||
"action": {
|
||||
"targetServers": ["premium-server-1", "premium-server-2"],
|
||||
"strategy": "least-connections",
|
||||
"qos": {
|
||||
"maxResponseTime": "1s",
|
||||
"priority": "high"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "high-volume-requests",
|
||||
"priority": 2,
|
||||
"condition": "request.size > 10MB",
|
||||
"action": {
|
||||
"targetServers": ["high-capacity-server"],
|
||||
"strategy": "single",
|
||||
"timeout": "60s"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "batch-processing",
|
||||
"priority": 3,
|
||||
"condition": "request.type === 'batch'",
|
||||
"action": {
|
||||
"targetServers": ["batch-server-1", "batch-server-2"],
|
||||
"strategy": "queue-based",
|
||||
"queueConfig": {
|
||||
"maxSize": 1000,
|
||||
"timeout": "5m"
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### JavaScript 路由函数
|
||||
|
||||
```javascript
|
||||
// 自定义路由函数
|
||||
function customRouting(request, servers, metrics) {
|
||||
const { user, content, timestamp } = request;
|
||||
|
||||
// 工作时间优先使用高性能服务器
|
||||
const isBusinessHours =
|
||||
new Date(timestamp).getHours() >= 9 && new Date(timestamp).getHours() <= 17;
|
||||
|
||||
if (isBusinessHours && user.priority === 'high') {
|
||||
return servers.filter((s) => s.tags.includes('high-performance'));
|
||||
}
|
||||
|
||||
// 基于内容类型的特殊路由
|
||||
if (content.type === 'code-analysis') {
|
||||
return servers.filter((s) => s.capabilities.includes('code-analysis'));
|
||||
}
|
||||
|
||||
// 默认负载均衡
|
||||
return servers.sort((a, b) => a.currentLoad - b.currentLoad);
|
||||
}
|
||||
```
|
||||
|
||||
## 路由优化
|
||||
|
||||
### 机器学习优化
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "ml-optimized",
|
||||
"mlConfig": {
|
||||
"algorithm": "reinforcement-learning",
|
||||
"rewardFunction": "response_time_weighted",
|
||||
"trainingData": {
|
||||
"features": [
|
||||
"server_load",
|
||||
"response_time_history",
|
||||
"request_complexity",
|
||||
"user_pattern",
|
||||
"time_of_day"
|
||||
],
|
||||
"targetMetric": "overall_satisfaction"
|
||||
},
|
||||
"updateFrequency": "hourly",
|
||||
"explorationRate": 0.1
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### 缓存感知路由
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"strategy": "cache-aware",
|
||||
"caching": {
|
||||
"enabled": true,
|
||||
"levels": [
|
||||
{
|
||||
"type": "local",
|
||||
"ttl": 300,
|
||||
"maxSize": "100MB"
|
||||
},
|
||||
{
|
||||
"type": "distributed",
|
||||
"provider": "redis",
|
||||
"ttl": 3600,
|
||||
"maxSize": "1GB"
|
||||
}
|
||||
],
|
||||
"routing": {
|
||||
"preferCachedServers": true,
|
||||
"cacheHitBonus": 0.3,
|
||||
"cacheMissThreshold": 0.8
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
``` */}
|
||||
|
||||
## 故障排除
|
||||
|
||||
### 路由调试
|
||||
<AccordionGroup>
|
||||
<Accordion title="数据库连接问题">
|
||||
**症状：**
|
||||
- 智能路由不可用
|
||||
- 数据库连接错误
|
||||
- 嵌入存储失败
|
||||
|
||||
```bash
|
||||
# 调试特定请求的路由决策
|
||||
curl -X POST http://localhost:3000/api/routing/debug \
|
||||
-H "Content-Type: application/json" \
|
||||
-H "Authorization: Bearer $TOKEN" \
|
||||
-d '{
|
||||
"request": {
|
||||
"userId": "user123",
|
||||
"tool": "filesystem",
|
||||
"content": {"type": "read", "path": "/data/file.txt"}
|
||||
},
|
||||
"traceRoute": true
|
||||
}'
|
||||
```
|
||||
**解决方案：**
|
||||
1. 验证 PostgreSQL 是否正在运行
|
||||
2. 检查 DATABASE_URL 格式
|
||||
3. 确保安装了 pgvector 扩展
|
||||
4. 手动测试连接：
|
||||
```bash
|
||||
psql $DATABASE_URL -c "SELECT 1;"
|
||||
```
|
||||
|
||||
### 路由性能分析
|
||||
</Accordion>
|
||||
|
||||
```bash
|
||||
# 获取路由性能报告
|
||||
curl -X GET http://localhost:3000/api/routing/performance \
|
||||
-H "Authorization: Bearer $TOKEN" \
|
||||
-G -d "timeRange=1h" -d "detailed=true"
|
||||
```
|
||||
<Accordion title="嵌入服务问题">
|
||||
**症状：**
|
||||
- 工具索引失败
|
||||
- 查询处理错误
|
||||
- API 速率限制错误
|
||||
|
||||
### 常见问题
|
||||
**解决方案：**
|
||||
1. 验证 API 密钥有效性
|
||||
2. 检查网络连接
|
||||
3. 监控速率限制
|
||||
4. 测试嵌入服务：
|
||||
```bash
|
||||
curl -X POST https://api.openai.com/v1/embeddings \
|
||||
-H "Authorization: Bearer $OPENAI_API_KEY" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{"input": "test", "model": "text-embedding-3-small"}'
|
||||
```
|
||||
|
||||
1. **不均匀的负载分布**
|
||||
</Accordion>
|
||||
|
||||
- 检查服务器权重配置
|
||||
- 验证健康检查设置
|
||||
- 分析请求模式
|
||||
<Accordion title="搜索结果不佳">
|
||||
**症状：**
|
||||
- 返回不相关的工具
|
||||
- 相关性得分低
|
||||
- 缺少预期的工具
|
||||
|
||||
2. **频繁的故障转移**
|
||||
**解决方案：**
|
||||
1. 调整相似性阈值
|
||||
2. 使用更好的描述重新索引工具
|
||||
3. 使用更具体的查询
|
||||
4. 检查工具元数据质量
|
||||
```bash
|
||||
# 重新索引所有工具
|
||||
curl -X POST http://localhost:3000/api/smart-routing/reindex \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN"
|
||||
```
|
||||
|
||||
- 调整健康检查阈值
|
||||
- 检查网络连接稳定性
|
||||
- 优化服务器资源
|
||||
</Accordion>
|
||||
|
||||
3. **路由延迟过高**
|
||||
- 简化路由规则
|
||||
- 优化路由算法
|
||||
- 使用缓存加速决策
|
||||
<Accordion title="性能问题">
|
||||
**症状：**
|
||||
- 查询响应缓慢
|
||||
- 数据库负载高
|
||||
- 内存使用激增
|
||||
|
||||
有关更多信息，请参阅 [监控](/zh/features/monitoring) 和 [服务器管理](/zh/features/server-management) 文档。
|
||||
**解决方案：**
|
||||
1. 优化数据库配置
|
||||
2. 增加缓存大小
|
||||
3. 减少批处理大小
|
||||
4. 监控系统资源
|
||||
```bash
|
||||
# 检查系统性能
|
||||
curl http://localhost:3000/api/smart-routing/performance \
|
||||
-H "Authorization: Bearer YOUR_JWT_TOKEN"
|
||||
```
|
||||
|
||||
</Accordion>
|
||||
</AccordionGroup>
|
||||
|
||||
## 最佳实践
|
||||
|
||||
### 查询编写
|
||||
|
||||
<Tip>
|
||||
**要具体描述**：在查询中使用具体、描述性的语言以获得更好的工具匹配。
|
||||
</Tip>
|
||||
|
||||
<Tip>
|
||||
**包含上下文**：提供有关您的任务或领域的相关上下文以获得更准确的结果。
|
||||
</Tip>
|
||||
|
||||
<Tip>**使用自然语言**：像向人类描述任务一样编写查询。</Tip>
|
||||
|
||||
### 工具描述
|
||||
|
||||
<Warning>
|
||||
**质量元数据**：确保 MCP 服务器提供高质量的工具描述和元数据。
|
||||
</Warning>
|
||||
|
||||
<Warning>**定期更新**：随着功能的发展保持工具描述的最新状态。</Warning>
|
||||
|
||||
<Warning>
|
||||
**一致的命名**：在工具和服务器中使用一致的命名约定。
|
||||
</Warning>
|
||||
|
||||
### 系统维护
|
||||
|
||||
<Info>**定期重新索引**：定期重新索引工具以确保嵌入质量。</Info>
|
||||
|
||||
<Info>**监控性能**：跟踪查询模式并根据使用情况进行优化。</Info>
|
||||
|
||||
<Info>
|
||||
**更新模型**：随着新嵌入模型的出现，考虑更新到更新的模型。
|
||||
</Info>
|
||||
|
||||
## 下一步
|
||||
|
||||
<CardGroup cols={2}>
|
||||
<Card title="身份验证" icon="shield" href="/zh/features/authentication">
|
||||
用户管理和访问控制
|
||||
</Card>
|
||||
<Card title="监控" icon="chart-line" href="/zh/features/monitoring">
|
||||
系统监控和分析
|
||||
</Card>
|
||||
<Card title="API 参考" icon="code" href="/zh/api-reference/smart-routing">
|
||||
完整的智能路由 API 文档
|
||||
</Card>
|
||||
<Card title="配置" icon="cog" href="/zh/configuration/environment-variables">
|
||||
高级配置选项
|
||||
</Card>
|
||||
</CardGroup>
|
||||
|
||||
Reference in New Issue
Block a user