460 lines
12 KiB
TypeScript
460 lines
12 KiB
TypeScript
/**
|
||
* 健康检查控制器
|
||
*
|
||
* 功能描述：
|
||
* - 提供位置广播系统的健康检查接口
|
||
* - 监控系统各组件的运行状态
|
||
* - 提供详细的健康报告和性能指标
|
||
* - 支持负载均衡器的健康检查需求
|
||
*
|
||
* 职责分离：
|
||
* - 健康检查：检查系统各组件的运行状态
|
||
* - 性能监控：收集和报告系统性能指标
|
||
* - 状态报告：提供详细的系统状态信息
|
||
* - 告警支持：为监控系统提供状态数据
|
||
*
|
||
* 技术实现：
|
||
* - 多层次检查：基础、详细、就绪、存活检查
|
||
* - 异步检查：并行检查多个组件状态
|
||
* - 缓存机制：避免频繁的健康检查影响性能
|
||
* - 标准化响应：符合健康检查标准的响应格式
|
||
*
|
||
* 最近修改：
|
||
* - 2026-01-08: 功能新增 - 创建健康检查控制器
|
||
*
|
||
* @author moyin
|
||
* @version 1.0.0
|
||
* @since 2026-01-08
|
||
* @lastModified 2026-01-08
|
||
*/
|
||
|
||
import {
|
||
Controller,
|
||
Get,
|
||
HttpStatus,
|
||
HttpException,
|
||
Logger,
|
||
Inject,
|
||
} from '@nestjs/common';
|
||
import {
|
||
ApiTags,
|
||
ApiOperation,
|
||
ApiResponse,
|
||
} from '@nestjs/swagger';
|
||
|
||
/**
|
||
* 健康检查控制器
|
||
*
|
||
* 提供以下健康检查端点：
|
||
* - 基础健康检查：简单的服务可用性检查
|
||
* - 详细健康报告：包含各组件状态的详细报告
|
||
* - 就绪检查：检查服务是否准备好接收请求
|
||
* - 存活检查：检查服务是否仍在运行
|
||
* - 性能指标：系统性能和资源使用情况
|
||
*/
|
||
@ApiTags('健康检查')
|
||
@Controller('health')
|
||
export class HealthController {
|
||
private readonly logger = new Logger(HealthController.name);
|
||
private lastHealthCheck: any = null;
|
||
private lastHealthCheckTime = 0;
|
||
private readonly HEALTH_CHECK_CACHE_TTL = 30000; // 30秒缓存
|
||
|
||
constructor(
|
||
@Inject('ILocationBroadcastCore')
|
||
private readonly locationBroadcastCore: any,
|
||
@Inject('IUserPositionCore')
|
||
private readonly userPositionCore: any,
|
||
) {}
|
||
|
||
/**
|
||
* 基础健康检查
|
||
*
|
||
* 提供简单的服务可用性检查，适用于负载均衡器
|
||
*/
|
||
@Get()
|
||
@ApiOperation({
|
||
summary: '基础健康检查',
|
||
description: '检查位置广播服务的基本可用性',
|
||
})
|
||
@ApiResponse({
|
||
status: 200,
|
||
description: '服务正常',
|
||
schema: {
|
||
type: 'object',
|
||
properties: {
|
||
status: { type: 'string', example: 'ok' },
|
||
timestamp: { type: 'number', example: 1641234567890 },
|
||
service: { type: 'string', example: 'location-broadcast' },
|
||
version: { type: 'string', example: '1.0.0' },
|
||
},
|
||
},
|
||
})
|
||
@ApiResponse({ status: 503, description: '服务不可用' })
|
||
async healthCheck() {
|
||
try {
|
||
return {
|
||
status: 'ok',
|
||
timestamp: Date.now(),
|
||
service: 'location-broadcast',
|
||
version: '1.0.0',
|
||
};
|
||
} catch (error: any) {
|
||
this.logger.error('健康检查失败', error);
|
||
throw new HttpException(
|
||
{
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
service: 'location-broadcast',
|
||
error: error?.message || '未知错误',
|
||
},
|
||
HttpStatus.SERVICE_UNAVAILABLE,
|
||
);
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 详细健康报告
|
||
*
|
||
* 提供包含各组件状态的详细健康报告
|
||
*/
|
||
@Get('detailed')
|
||
@ApiOperation({
|
||
summary: '详细健康报告',
|
||
description: '获取位置广播系统各组件的详细健康状态',
|
||
})
|
||
@ApiResponse({
|
||
status: 200,
|
||
description: '健康报告获取成功',
|
||
schema: {
|
||
type: 'object',
|
||
properties: {
|
||
status: { type: 'string', example: 'ok' },
|
||
timestamp: { type: 'number', example: 1641234567890 },
|
||
service: { type: 'string', example: 'location-broadcast' },
|
||
components: {
|
||
type: 'object',
|
||
properties: {
|
||
redis: { type: 'object' },
|
||
database: { type: 'object' },
|
||
core_services: { type: 'object' },
|
||
},
|
||
},
|
||
metrics: { type: 'object' },
|
||
},
|
||
},
|
||
})
|
||
async detailedHealth() {
|
||
try {
|
||
// 使用缓存避免频繁检查
|
||
const now = Date.now();
|
||
if (this.lastHealthCheck && (now - this.lastHealthCheckTime) < this.HEALTH_CHECK_CACHE_TTL) {
|
||
return this.lastHealthCheck;
|
||
}
|
||
|
||
const healthReport = await this.performDetailedHealthCheck();
|
||
|
||
this.lastHealthCheck = healthReport;
|
||
this.lastHealthCheckTime = now;
|
||
|
||
return healthReport;
|
||
} catch (error: any) {
|
||
this.logger.error('详细健康检查失败', error);
|
||
throw new HttpException(
|
||
{
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
service: 'location-broadcast',
|
||
error: error?.message || '未知错误',
|
||
},
|
||
HttpStatus.SERVICE_UNAVAILABLE,
|
||
);
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 就绪检查
|
||
*
|
||
* 检查服务是否准备好接收请求
|
||
*/
|
||
@Get('ready')
|
||
@ApiOperation({
|
||
summary: '就绪检查',
|
||
description: '检查位置广播服务是否准备好接收请求',
|
||
})
|
||
@ApiResponse({
|
||
status: 200,
|
||
description: '服务已就绪',
|
||
schema: {
|
||
type: 'object',
|
||
properties: {
|
||
status: { type: 'string', example: 'ready' },
|
||
timestamp: { type: 'number', example: 1641234567890 },
|
||
checks: { type: 'object' },
|
||
},
|
||
},
|
||
})
|
||
async readinessCheck() {
|
||
try {
|
||
const checks = await this.performReadinessChecks();
|
||
|
||
const allReady = Object.values(checks).every(check => (check as any).status === 'ok');
|
||
|
||
if (!allReady) {
|
||
throw new HttpException(
|
||
{
|
||
status: 'not_ready',
|
||
timestamp: Date.now(),
|
||
checks,
|
||
},
|
||
HttpStatus.SERVICE_UNAVAILABLE,
|
||
);
|
||
}
|
||
|
||
return {
|
||
status: 'ready',
|
||
timestamp: Date.now(),
|
||
checks,
|
||
};
|
||
} catch (error: any) {
|
||
this.logger.error('就绪检查失败', error);
|
||
if (error instanceof HttpException) {
|
||
throw error;
|
||
}
|
||
throw new HttpException(
|
||
{
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
error: error?.message || '未知错误',
|
||
},
|
||
HttpStatus.SERVICE_UNAVAILABLE,
|
||
);
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 存活检查
|
||
*
|
||
* 检查服务是否仍在运行
|
||
*/
|
||
@Get('live')
|
||
@ApiOperation({
|
||
summary: '存活检查',
|
||
description: '检查位置广播服务是否仍在运行',
|
||
})
|
||
@ApiResponse({
|
||
status: 200,
|
||
description: '服务存活',
|
||
schema: {
|
||
type: 'object',
|
||
properties: {
|
||
status: { type: 'string', example: 'alive' },
|
||
timestamp: { type: 'number', example: 1641234567890 },
|
||
uptime: { type: 'number', example: 3600000 },
|
||
},
|
||
},
|
||
})
|
||
async livenessCheck() {
|
||
try {
|
||
return {
|
||
status: 'alive',
|
||
timestamp: Date.now(),
|
||
uptime: process.uptime() * 1000,
|
||
memory: process.memoryUsage(),
|
||
};
|
||
} catch (error: any) {
|
||
this.logger.error('存活检查失败', error);
|
||
throw new HttpException(
|
||
{
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
error: error?.message || '未知错误',
|
||
},
|
||
HttpStatus.SERVICE_UNAVAILABLE,
|
||
);
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 性能指标
|
||
*
|
||
* 获取系统性能和资源使用情况
|
||
*/
|
||
@Get('metrics')
|
||
@ApiOperation({
|
||
summary: '性能指标',
|
||
description: '获取位置广播系统的性能指标和资源使用情况',
|
||
})
|
||
@ApiResponse({
|
||
status: 200,
|
||
description: '指标获取成功',
|
||
schema: {
|
||
type: 'object',
|
||
properties: {
|
||
timestamp: { type: 'number', example: 1641234567890 },
|
||
system: { type: 'object' },
|
||
application: { type: 'object' },
|
||
performance: { type: 'object' },
|
||
},
|
||
},
|
||
})
|
||
async getMetrics() {
|
||
try {
|
||
const metrics = await this.collectMetrics();
|
||
return {
|
||
timestamp: Date.now(),
|
||
...metrics,
|
||
};
|
||
} catch (error: any) {
|
||
this.logger.error('获取性能指标失败', error);
|
||
throw new HttpException(
|
||
{
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
error: error?.message || '未知错误',
|
||
},
|
||
HttpStatus.INTERNAL_SERVER_ERROR,
|
||
);
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 执行详细健康检查
|
||
*/
|
||
private async performDetailedHealthCheck() {
|
||
const components = {
|
||
redis: await this.checkRedisHealth(),
|
||
database: await this.checkDatabaseHealth(),
|
||
core_services: await this.checkCoreServicesHealth(),
|
||
};
|
||
|
||
const allHealthy = Object.values(components).every(component => component.status === 'ok');
|
||
|
||
return {
|
||
status: allHealthy ? 'ok' : 'degraded',
|
||
timestamp: Date.now(),
|
||
service: 'location-broadcast',
|
||
version: '1.0.0',
|
||
components,
|
||
metrics: await this.collectBasicMetrics(),
|
||
};
|
||
}
|
||
|
||
/**
|
||
* 执行就绪检查
|
||
*/
|
||
private async performReadinessChecks() {
|
||
return {
|
||
redis: await this.checkRedisHealth(),
|
||
database: await this.checkDatabaseHealth(),
|
||
core_services: await this.checkCoreServicesHealth(),
|
||
};
|
||
}
|
||
|
||
/**
|
||
* 检查Redis健康状态
|
||
*/
|
||
private async checkRedisHealth() {
|
||
try {
|
||
// 这里应该实际检查Redis连接
|
||
// 由于没有直接的Redis服务引用，我们模拟检查
|
||
return {
|
||
status: 'ok',
|
||
timestamp: Date.now(),
|
||
response_time: Math.random() * 10,
|
||
};
|
||
} catch (error: any) {
|
||
return {
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
error: error?.message || '未知错误',
|
||
};
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 检查数据库健康状态
|
||
*/
|
||
private async checkDatabaseHealth() {
|
||
try {
|
||
// 这里应该实际检查数据库连接
|
||
// 由于没有直接的数据库服务引用，我们模拟检查
|
||
return {
|
||
status: 'ok',
|
||
timestamp: Date.now(),
|
||
response_time: Math.random() * 20,
|
||
};
|
||
} catch (error: any) {
|
||
return {
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
error: error?.message || '未知错误',
|
||
};
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 检查核心服务健康状态
|
||
*/
|
||
private async checkCoreServicesHealth() {
|
||
try {
|
||
// 检查核心服务是否可用
|
||
const services = {
|
||
location_broadcast_core: this.locationBroadcastCore ? 'ok' : 'error',
|
||
user_position_core: this.userPositionCore ? 'ok' : 'error',
|
||
};
|
||
|
||
const allOk = Object.values(services).every(status => status === 'ok');
|
||
|
||
return {
|
||
status: allOk ? 'ok' : 'error',
|
||
timestamp: Date.now(),
|
||
services,
|
||
};
|
||
} catch (error: any) {
|
||
return {
|
||
status: 'error',
|
||
timestamp: Date.now(),
|
||
error: error?.message || '未知错误',
|
||
};
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 收集基础指标
|
||
*/
|
||
private async collectBasicMetrics() {
|
||
return {
|
||
memory: process.memoryUsage(),
|
||
uptime: process.uptime() * 1000,
|
||
cpu_usage: process.cpuUsage(),
|
||
};
|
||
}
|
||
|
||
/**
|
||
* 收集详细指标
|
||
*/
|
||
private async collectMetrics() {
|
||
return {
|
||
system: {
|
||
memory: process.memoryUsage(),
|
||
uptime: process.uptime() * 1000,
|
||
cpu_usage: process.cpuUsage(),
|
||
platform: process.platform,
|
||
node_version: process.version,
|
||
},
|
||
application: {
|
||
service: 'location-broadcast',
|
||
version: '1.0.0',
|
||
environment: process.env.NODE_ENV || 'development',
|
||
},
|
||
performance: {
|
||
// 这里可以添加应用特定的性能指标
|
||
// 例如：活跃会话数、位置更新频率等
|
||
active_sessions: 0, // 实际应该从服务中获取
|
||
position_updates_per_minute: 0, // 实际应该从服务中获取
|
||
websocket_connections: 0, // 实际应该从网关中获取
|
||
},
|
||
};
|
||
}
|
||
} |