node.js - NodeJS : how to run spawn in parallel properly?

Question

我正在运行一个 for 循环，其中执行以下内容:

let perform_vrp = function() {
    //..
    perform_tsp();
}

let perform_tsp = function() {
  //..

  const pyProg = spawn('python3', [process.env.PWD + '/server/vrp_solver/tsp_solver.py', '/../../route_data/' + depot.city + '/' + moment(route.date_driven).format('Y-MM-DD'), 'morning',route.name]);

  winston.info('Solving the TSP for %s...', route.name);

  pyProg.stdout.on('data', function (data) {

    let result_string = data.toString();
    winston.info('Route result for %s is: %s', route.name, result_string);

    let result_array = eval(result_string); 
    //...
  });
}

它基本上为 for 循环中的每个项目调用一个 python 脚本。

但是，当一个脚本完成后，它也会完成所有其他脚本，并继续使用 stdout 对 for 循环中的所有项目使用相同的“数据”。

如何防止这种情况发生并让 stdout 等待正确的子进程完成？

更新:

以上是由以下命令触发的:

const winston = require('../../server/winston');
const perform_vrp = require('../../server/modules/vrp');
const moment = require('moment');
const Utils = require('../modules/utils');
let utils = new Utils();

let do_vrp = async function(next_delivery_date, cities) {
  winston.info('Generating routes for %s', next_delivery_date);
  for (let index in cities) {
    if (cities.hasOwnProperty(index)) {
      let city = cities[index];
      winston.info('Generating route for %s on %s', city, next_delivery_date);
      await perform_vrp(next_delivery_date,[city]);
      await utils.sleep(60000);
    }
  }
};


let process_routes = async function() {

  //...

  let morning_cities = ['Boston','Chicago'];
  await do_vrp(next_delivery_date.format('YYYY-MM-DD'), morning_cities);
};

process_routes();

Answer 1

问题是您在 for 循环内使用 wait 来等待脚本完成，但没有将脚本的执行包装在 Promise 中。

在perform_tsp中

扭曲 Promise 中的生成和事件，当 data 事件触发时解析并使用接收到的数据进行解析，如下所示:

  let perform_tsp = function () {
  //..
  return new Promise((resolve, reject) => {
    const pyProg = spawn('python3', [process.env.PWD + '/server/vrp_solver/tsp_solver.py', '/../../route_data/' + depot.city + '/' + moment(route.date_driven).format('Y-MM-DD'), 'morning', route.name]);
    winston.info('Solving the TSP for %s...', route.name);
    pyProg.stdout.on('data', function (data) {
      resolve(data);
    })
  });
}

在perform_vrp中

只需返回从perform_tsp返回的promise，这样你就可以在for循环中等待它:

let perform_vrp = function () {
  //..
  return perform_tsp();
}

在do_vrp中

接收已解决的 Promise 的数据，即生成的数据，之后您可以执行 data 事件回调中的逻辑:

 let do_vrp = async function (next_delivery_date, cities) {
  winston.info('Generating routes for %s', next_delivery_date);
  for (let index in cities) {
    if (cities.hasOwnProperty(index)) {
      let city = cities[index];
      winston.info('Generating route for %s on %s', city, next_delivery_date);
      let data = await perform_vrp(next_delivery_date, [city]);
      let result_string = data.toString();
      winston.info('Route result for %s is: %s', route.name, result_string);
      let result_array = eval(result_string);
      await utils.sleep(60000);
    }
  }
};

如果您想使用 Promise.all api 并行重新生成生成，则先前的结构将串行运行生成。

  let do_vrp = async function (next_delivery_date, cities) {
  winston.info('Generating routes for %s', next_delivery_date);

  const promises = cities.map((city) => {
    return perform_vrp(next_delivery_date, [city]);
  });

  const results = await Promise.all(promises);
  results.forEach((data) => {
    winston.info('Generating route for %s on %s', city, next_delivery_date);
    let result_string = data.toString();
    winston.info('Route result for %s is: %s', route.name, result_string);
    let result_array = eval(result_string);
  });
};