qinglong/back/app.ts

import 'reflect-metadata';
import cluster, { type Worker } from 'cluster';
import type express from 'express';
import { Container } from 'typedi';
import config from './config';
import Logger from './loaders/logger';
import { errStack } from './shared/errors';
import { initializeLanguageSync } from './shared/languageSync';
import { type GrpcServerService } from './services/grpc';
import { type HttpServerService } from './services/http';

interface WorkerMetadata {
  id: number;
  pid: number;
  serviceType: string;
  startTime: Date;
}

class Application {
  private httpServerService?: HttpServerService;
  private grpcServerService?: GrpcServerService;
  private isShuttingDown = false;
  private workerMetadataMap = new Map<number, WorkerMetadata>();
  private httpWorker?: Worker;

  async start() {
    try {
      if (cluster.isPrimary) {
        await this.initializeDatabase();
      }
      await initializeLanguageSync();
      if (cluster.isPrimary) {
        this.startMasterProcess();
      } else {
        await this.startWorkerProcess();
      }
    } catch (error) {
      Logger.error(`Failed to start application:\n${errStack(error)}`);
      process.exit(1);
    }
  }

  private startMasterProcess() {
    const readyTimeout =
      process.arch === 'arm' &&
      Number(
        (process.config.variables as { arm_version?: string | number }).arm_version,
      ) === 6
        ? 120000
        : 30000;
    // Fork gRPC worker first and wait for it to be ready
    const grpcWorker = this.forkWorker('grpc');
    
    // Wait for gRPC worker to signal it's ready before starting HTTP worker
    this.waitForWorkerReady(grpcWorker, readyTimeout)
      .then(() => {
        Logger.info('[boot] gRPC worker is ready, starting HTTP worker');
        this.httpWorker = this.forkWorker('http');
      })
      .catch((error) => {
        Logger.error(`[boot] Failed to wait for gRPC worker:\n${errStack(error)}`);
        process.exit(1);
      });

    cluster.on('exit', (worker, code, signal) => {
      const metadata = this.workerMetadataMap.get(worker.id);
      if (metadata) {
        if (!this.isShuttingDown) {
          Logger.error(
            `${metadata.serviceType} worker ${worker.process.pid} died (${signal || code
            }). Restarting...`,
          );
          // If gRPC worker died, restart it and wait for it to be ready
          if (metadata.serviceType === 'grpc') {
            try {
              this.httpWorker?.send('scheduler-unavailable');
            } catch (error) {
              Logger.warn('Unable to notify HTTP worker of scheduler exit');
            }
            const newGrpcWorker = this.forkWorker('grpc');
            this.waitForWorkerReady(newGrpcWorker, readyTimeout)
              .then(() => {
                Logger.info('gRPC worker restarted and ready');
                // Re-register cron jobs by notifying the HTTP worker
                if (this.httpWorker) {
                  try {
                    this.httpWorker.send('reregister-crons');
                    Logger.info('Sent reregister-crons message to HTTP worker');
                  } catch (error) {
                    Logger.error(`Failed to send reregister-crons message:\n${errStack(error)}`);
                  }
                }
              })
              .catch((error) => {
                Logger.error(`Failed to restart gRPC worker:\n${errStack(error)}`);
                process.exit(1);
              });
          } else {
            // For HTTP worker, just restart it
            const newWorker = this.forkWorker(metadata.serviceType);
            this.httpWorker = newWorker;
            Logger.info(`Restarted ${metadata.serviceType} worker (PID: ${newWorker.process.pid})`);
          }
        }

        this.workerMetadataMap.delete(worker.id);
      }
    });

    this.setupMasterShutdown();
  }

  private waitForWorkerReady(worker: Worker, timeoutMs: number): Promise<void> {
    return new Promise<void>((resolve, reject) => {
      const messageHandler = (msg: any) => {
        if (msg === 'ready') {
          worker.removeListener('message', messageHandler);
          clearTimeout(timeoutId);
          resolve();
        }
      };
      worker.on('message', messageHandler);
      
      // Timeout after specified milliseconds
      const timeoutId = setTimeout(() => {
        worker.removeListener('message', messageHandler);
        reject(new Error(`Worker failed to start within ${timeoutMs / 1000} seconds`));
      }, timeoutMs);
    });
  }

  private forkWorker(serviceType: string): Worker {
    const workerEnv: NodeJS.ProcessEnv = { SERVICE_TYPE: serviceType };
    // PM2's fork launcher is inherited by our own cluster workers. Their APM
    // messages go to this primary, not PM2, and duplicate its sampling work.
    // Keep primary monitoring and allow restoring the inherited worker APM.
    if (process.env.pm_id !== undefined && process.env.QL_WORKER_APM !== 'true') {
      workerEnv.pmx = 'false';
    }
    const worker = cluster.fork(workerEnv);

    this.workerMetadataMap.set(worker.id, {
      id: worker.id,
      pid: worker.process.pid!,
      serviceType,
      startTime: new Date(),
    });

    return worker;
  }

  private async initializeDatabase() {
    const { runStartupProcess } = await import('./shared/startupProcess');
    await runStartupProcess(require.resolve('./bootstrap/database'));
  }

  private async setupMiddlewares(): Promise<express.Application> {
    const { createHttpApplication } = await import('./loaders/httpApplication');
    return createHttpApplication();
  }

  private setupMasterShutdown() {
    const shutdown = async () => {
      if (this.isShuttingDown) return;
      this.isShuttingDown = true;

      const workers = Object.values(cluster.workers || {});
      const workerPromises: Promise<void>[] = [];

      workers.forEach((worker) => {
        if (worker) {
          const exitPromise = new Promise<void>((resolve) => {
            worker.once('exit', () => {
              Logger.info(`Worker ${worker.process.pid} exited`);
              resolve();
            });

            try {
              worker.send('shutdown');
            } catch (error) {
              Logger.warn(`Failed to send shutdown to worker ${worker.process.pid}:\n${errStack(error)}`);
            }
          });

          workerPromises.push(exitPromise);
        }
      });

      try {
        await Promise.race([
          Promise.all(workerPromises),
          new Promise<void>((resolve) => {
            setTimeout(() => {
              Logger.warn('Worker shutdown timeout reached');
              resolve();
            }, 10000);
          }),
        ]);
        process.exit(0);
      } catch (error) {
        Logger.error(`Error during worker shutdown:\n${errStack(error)}`);
        process.exit(1);
      }
    };

    process.on('SIGTERM', shutdown);
    process.on('SIGINT', shutdown);
  }

  private async startWorkerProcess() {
    const serviceType = process.env.SERVICE_TYPE;
    if (!serviceType || !['http', 'grpc'].includes(serviceType)) {
      Logger.error('[boot] Invalid SERVICE_TYPE:', serviceType);
      process.exit(1);
    }

    Logger.info(`[boot] ${serviceType} worker started (PID: ${process.pid})`);

    try {
      if (serviceType === 'http') {
        await this.startHttpService();
      } else {
        await this.startGrpcService();
      }

      process.send?.('ready');
    } catch (error) {
      Logger.error(`[boot] ${serviceType} worker failed:\n${errStack(error)}`);
      process.exit(1);
    }
  }

  private async startHttpService() {
    // 在导入任何 gRPC 客户端模块之前初始化 mTLS 证书
    const { initGrpcCerts } = await import('./config/grpcCerts');
    await initGrpcCerts();

    const app = await this.setupMiddlewares();

    const { HttpServerService } = await import('./services/http');
    this.httpServerService = Container.get(HttpServerService);

    const appLoader = await import('./loaders/app');
    await appLoader.default({ app });

    const server = await this.httpServerService.initialize(
      app,
      config.port,
    );

    const serverLoader = await import('./loaders/server');
    await (serverLoader.default as any)({ server });
    this.setupWorkerShutdown('http');
  }

  private async startGrpcService() {
    const { GrpcServerService } = await import('./services/grpc');
    this.grpcServerService = Container.get(GrpcServerService);

    await this.grpcServerService.initialize();
    this.setupWorkerShutdown('grpc');
  }

  private setupWorkerShutdown(serviceType: string) {
    process.on('message', async (msg) => {
      if (msg === 'shutdown') {
        this.gracefulShutdown(serviceType);
      } else if (serviceType === 'http' &&
        (msg === 'reregister-crons' || msg === 'scheduler-unavailable')) {
        const { default: cronClient } = await import('./schedule/client');
        cronClient.readiness.invalidate();
      }
    });

    const shutdown = () => this.gracefulShutdown(serviceType);
    process.on('SIGTERM', shutdown);
    process.on('SIGINT', shutdown);
  }

  private async gracefulShutdown(serviceType: string) {
    if (this.isShuttingDown) return;
    this.isShuttingDown = true;

    try {
      if (serviceType === 'http') {
        await this.httpServerService?.shutdown();
      } else {
        await this.grpcServerService?.shutdown();
      }
      process.exit(0);
    } catch (error) {
      Logger.error(`[${serviceType}] Error during shutdown:\n${errStack(error)}`);
      process.exit(1);
    }
  }
}

const app = new Application();
app.start().catch((error) => {
  Logger.error(`🙅‍♀️ Application failed to start:\n${errStack(error)}`);
  process.exit(1);
});