如何使用集群提高NodeJS服务器的吞吐量？

Question

我有一个NodeJS服务器(Express)，我使用nodeJs站点上的集群模块示例将请求扩展到多个处理器。

if (cluster.isMaster) {
  for (var i = 0; i < numCPUs; i++) {
    cluster.fork();
  };
  cluster.on('exit', function(worker, code, signal) {
    console.log('worker ' + worker.process.pid + ' died');
    cluster.fork();
  });
} else {  
  server.listen(app.get('port'), function(){
    console.log('HTTP server on port ' + app.get('port') + ' - running as ' + app.settings.env);
  });

  // setup socket.io communication
  io.sockets.on('connection', require('./app/sockets'));
  io.sockets.on('connection', require('./app/downloadSockets'));
}

问题是，围城的基准告诉我，点击次数没有增加。这是围困的产物：

$ siege -c100 192.168.111.1:42424  -t10S
** SIEGE 3.0.5
** Preparing 100 concurrent users for battle.
The server is now under siege...
Lifting the server siege...      done.

Transactions:               1892 hits
Availability:             100.00 %
Elapsed time:              10.01 secs
Data transferred:           9.36 MB
Response time:              0.01 secs
Transaction rate:         189.01 trans/sec
Throughput:             0.93 MB/sec
Concurrency:                1.58
Successful transactions:        1892
Failed transactions:               0
Longest transaction:            0.05
Shortest transaction:           0.00

聚类后：

$ siege -c100 192.168.111.1:42424  -t10S
** SIEGE 3.0.5
** Preparing 100 concurrent users for battle.
The server is now under siege...
Lifting the server siege...      done.

Transactions:               1884 hits
Availability:             100.00 %
Elapsed time:               9.52 secs
Data transferred:           9.32 MB
Response time:              0.01 secs
Transaction rate:         197.90 trans/sec
Throughput:             0.98 MB/sec
Concurrency:                1.72
Successful transactions:        1884
Failed transactions:               0
Longest transaction:            0.07
Shortest transaction:           0.00

这是否意味着我的服务器已经获得了最大的单服务器吞吐量，可能是因为它是一台本地机器，或者它无法获得4个处理器，因为运行的进程太多，我不确定。

如何使用集群模块来增加节流，以及为什么我的当前代码不成功？此外，我还检查了它是否确实创建了服务器的4个实例，即cluster.fork工作。任何建议都是非常有用的。

Answer 1

这种效果是通过集群或增长并发查询来实现的(尝试将并发用户的数量增加到300-400)。或者是给你带来沉重负担的任务。让我们花费更多有趣的测试:将下载大约1MB的文件大小，另外，为了模拟复杂的操作，我们会延迟5毫秒和50毫秒。对于本地测试中的四个核心处理器，如下所示(分别用于正常和集群)：

$ siege -c100 http://localhost/images/image.jpg  -t10S

正常模式(5毫秒延迟)：

Lifting the server siege...      done.
Transactions:                    1170 hits
Availability:                 100.00 %
Elapsed time:                   9.10 secs
Data transferred:             800.79 MB
Response time:                  0.27 secs
Transaction rate:             128.57 trans/sec
Throughput:                    88.00 MB/sec
Concurrency:                   34.84
Successful transactions:        1170
Failed transactions:               0
Longest transaction:            0.95
Shortest transaction:           0.01

集群模式(延迟5毫秒)：

Lifting the server siege...      done.
Transactions:                    1596 hits
Availability:                 100.00 %
Elapsed time:                   9.04 secs
Data transferred:            1092.36 MB
Response time:                  0.06 secs
Transaction rate:             176.55 trans/sec
Throughput:                   120.84 MB/sec
Concurrency:                    9.81
Successful transactions:        1596
Failed transactions:               0
Longest transaction:            0.33
Shortest transaction:           0.00

正常模式(50毫秒延迟)：

Lifting the server siege...      done.
Transactions:                     100 hits
Availability:                 100.00 %
Elapsed time:                   9.63 secs
Data transferred:              68.44 MB
Response time:                  5.51 secs
Transaction rate:              10.38 trans/sec
Throughput:                     7.11 MB/sec
Concurrency:                   57.18
Successful transactions:         100
Failed transactions:               0
Longest transaction:            7.77
Shortest transaction:           5.14

集群模式(延迟50毫秒)：

Lifting the server siege...      done.
Transactions:                     614 hits
Availability:                 100.00 %
Elapsed time:                   9.24 secs
Data transferred:             420.25 MB
Response time:                  0.90 secs
Transaction rate:              66.45 trans/sec
Throughput:                    45.48 MB/sec
Concurrency:                   59.59
Successful transactions:         614
Failed transactions:               0
Longest transaction:            1.50
Shortest transaction:           0.50

Answer 2

在你的例子中，你并没有做任何事情。连接到mysql并运行一个繁重的查询，或者发出一个只需几秒钟的http请求。您注意到，最终，您将编写一些代码(或使用第三方库)来阻止事件循环。这是集群非常重要的时候，因为实际上每个处理器都有一个事件循环。如果一个查询很慢，而事件循环需要等待，它不会停止访问API/应用程序的新请求。

此外，如果计划使用或连接数据库或获取外部资源，则可能需要阅读连接池，特别是npm上的泛型池。