stock-bot/libs/data/questdb/src/health.ts

import { getLogger } from '@stock-bot/logger';
import type { HealthStatus, PerformanceMetrics, QueryResult } from './types';

// Interface to avoid circular dependency
interface QuestDBClientInterface {
  query<T = any>(sql: string, params?: any[]): Promise<QueryResult<T>>;
  isPgPoolHealthy(): boolean;
}

/**
 * QuestDB Health Monitor
 *
 * Monitors connection health, performance metrics, and provides
 * automatic recovery capabilities for the QuestDB client.
 */
export class QuestDBHealthMonitor {
  private readonly logger: ReturnType<typeof getLogger>;
  private healthCheckInterval: NodeJS.Timeout | null = null;
  private lastHealthCheck: Date | null = null;
  private performanceMetrics: PerformanceMetrics = {
    totalQueries: 0,
    successfulQueries: 0,
    failedQueries: 0,
    averageResponseTime: 0,
    lastQueryTime: null,
    connectionUptime: 0,
    memoryUsage: 0,
  };
  constructor(private readonly client: QuestDBClientInterface) {
    this.logger = getLogger('questdb-health-monitor');
  }

  /**
   * Start health monitoring
   */
  public startMonitoring(intervalMs: number = 30000): void {
    if (this.healthCheckInterval) {
      this.stopMonitoring();
    }

    this.logger.info(`Starting health monitoring with ${intervalMs}ms interval`);

    this.healthCheckInterval = setInterval(async () => {
      try {
        await this.performHealthCheck();
      } catch (error) {
        this.logger.error('Health check failed', error);
      }
    }, intervalMs);

    // Perform initial health check
    this.performHealthCheck().catch(error => {
      this.logger.error('Initial health check failed', error);
    });
  }

  /**
   * Stop health monitoring
   */
  public stopMonitoring(): void {
    if (this.healthCheckInterval) {
      clearInterval(this.healthCheckInterval);
      this.healthCheckInterval = null;
      this.logger.info('Health monitoring stopped');
    }
  }

  /**
   * Perform a health check
   */
  public async performHealthCheck(): Promise<HealthStatus> {
    const startTime = Date.now();

    try {
      // Test basic connectivity with a simple query
      await this.client.query('SELECT 1 as health_check');

      const responseTime = Date.now() - startTime;
      this.lastHealthCheck = new Date();

      const status: HealthStatus = {
        isHealthy: true,
        lastCheck: this.lastHealthCheck,
        responseTime,
        message: 'Connection healthy',
        details: {
          pgPool: this.client.isPgPoolHealthy(),
          httpEndpoint: true, // Will be implemented when HTTP client is added
          uptime: this.getUptime(),
        },
      };

      this.logger.debug('Health check passed', { responseTime });
      return status;
    } catch (error) {
      const responseTime = Date.now() - startTime;
      this.lastHealthCheck = new Date();

      const status: HealthStatus = {
        isHealthy: false,
        lastCheck: this.lastHealthCheck,
        responseTime,
        message: `Health check failed: ${error instanceof Error ? error.message : 'Unknown error'}`,
        error: error instanceof Error ? error : new Error('Unknown error'),
        details: {
          pgPool: false,
          httpEndpoint: false,
          uptime: this.getUptime(),
        },
      };

      this.logger.error('Health check failed', { error, responseTime });
      return status;
    }
  }

  /**
   * Get current health status
   */
  public async getHealthStatus(): Promise<HealthStatus> {
    if (!this.lastHealthCheck || Date.now() - this.lastHealthCheck.getTime() > 60000) {
      return await this.performHealthCheck();
    }

    // Return cached status if recent
    return {
      isHealthy: true,
      lastCheck: this.lastHealthCheck,
      responseTime: 0,
      message: 'Using cached health status',
      details: {
        pgPool: this.client.isPgPoolHealthy(),
        httpEndpoint: true,
        uptime: this.getUptime(),
      },
    };
  }

  /**
   * Record query performance metrics
   */
  public recordQuery(success: boolean, responseTime: number): void {
    this.performanceMetrics.totalQueries++;
    this.performanceMetrics.lastQueryTime = new Date();

    if (success) {
      this.performanceMetrics.successfulQueries++;
    } else {
      this.performanceMetrics.failedQueries++;
    }

    // Update rolling average response time
    const totalResponseTime =
      this.performanceMetrics.averageResponseTime * (this.performanceMetrics.totalQueries - 1) +
      responseTime;
    this.performanceMetrics.averageResponseTime =
      totalResponseTime / this.performanceMetrics.totalQueries;

    // Update memory usage
    this.performanceMetrics.memoryUsage = process.memoryUsage().heapUsed;
  }

  /**
   * Get performance metrics
   */
  public getPerformanceMetrics(): PerformanceMetrics {
    return { ...this.performanceMetrics };
  }

  /**
   * Get connection uptime in seconds
   */
  private getUptime(): number {
    return Math.floor(process.uptime());
  }

  /**
   * Reset performance metrics
   */
  public resetMetrics(): void {
    this.performanceMetrics = {
      totalQueries: 0,
      successfulQueries: 0,
      failedQueries: 0,
      averageResponseTime: 0,
      lastQueryTime: null,
      connectionUptime: this.getUptime(),
      memoryUsage: process.memoryUsage().heapUsed,
    };

    this.logger.debug('Performance metrics reset');
  }

  /**
   * Get health summary for monitoring dashboards
   */
  public async getHealthSummary(): Promise<{
    status: HealthStatus;
    metrics: PerformanceMetrics;
    recommendations: string[];
  }> {
    const status = await this.getHealthStatus();
    const metrics = this.getPerformanceMetrics();
    const recommendations: string[] = [];

    // Generate recommendations based on metrics
    if (metrics.failedQueries > metrics.successfulQueries * 0.1) {
      recommendations.push('High error rate detected - check query patterns');
    }

    if (metrics.averageResponseTime > 1000) {
      recommendations.push('High response times - consider query optimization');
    }

    if (metrics.memoryUsage > 100 * 1024 * 1024) {
      // 100MB
      recommendations.push('High memory usage - monitor for memory leaks');
    }

    return {
      status,
      metrics,
      recommendations,
    };
  }

  /**
   * Cleanup resources
   */
  public destroy(): void {
    this.stopMonitoring();
    this.logger.debug('Health monitor destroyed');
  }
}