Node.js

【Node.js】Node.js 应用性能监测与分析

Posted by 西维蜀黍 on 2018-11-12, Last Modified on 2021-09-21

监控指标（Metrics to watch）

Tick频率（Tick frequency） - number of ticks per time

Tick周期（Tick Duration） - the time one tick takes

Active Handles and Requests

Garbbage Collection activity

  • CPU 使用（CPU Usage）

  • 堆使用（Heap Usage）

  • 内存使用（Memory usage）

  • 吞吐量（Throughput）

  • 响应时间（Response Time）

开源项目

Olegas/node-event-loop-monitor

bretcope/node-gc-profiler

Bnoordhuis/node-heapdump

内存泄漏

性能优化的好处

  • 使用更少的内存
  • 使用更少的CPU资源
  • app运行更快

性能分析

场景1 - 长耗时的同步任务

当Node应用具有很高的CPU使用率（CPU usage）时，表明此时在进行大量的同步任务，这意味着Event Loop 中的异步回调很有可能会被延迟执行。

function fibonacci(n) {
    if(n==0 || n == 1)
        return n;
    return fibonacci(n-1) + fibonacci(n-2);
}

var timeoutScheduled = Date.now();

var result=fibonacci(45);
const delay = Date.now() - timeoutScheduled;
console.log("finish fibonacci");
console.log(`${delay}ms fibonacci`);


setTimeout(() => {
  const delay = Date.now() - timeoutScheduled;

  console.log('setTimeout');
  console.log(`${delay}ms setTimeout`);
}, 5);

setImmediate(() => {
	const delay = Date.now() - timeoutScheduled;

  console.log('setImmediate1');
  console.log(`${delay}ms setImmediate1`);
});

setImmediate(() => {
	const delay = Date.now() - timeoutScheduled;

  console.log('setImmediate2');
  console.log(`${delay}ms setImmediate2`);
});

process.nextTick(() => {
  const delay = Date.now() - timeoutScheduled;

  console.log('process.nextTick');
  console.log(`${delay}ms process.nextTick`);
});

执行结果

$ node main.js
finish fibonacci
12421ms fibonacci
process.nextTick
12424ms process.nextTick
setImmediate1
12425ms setImmediate1
setImmediate2
12425ms setImmediate2
setTimeout
12429ms setTimeout

分析

可以看到，同步任务fibonacci执行了12421ms，所有的异步回调都被阻塞了，直到这个同步任务完成后，才依次被执行。

同时，该node进程的CPU使用率为几乎100%，而整机的CPU的使用率却相对非常低（空闲率为80%），这是因为这个fibonacci同步任务仅仅在一个线程中执行，因此无法充分利用多核CPU的计算能力。

判断异步回调是否被阻塞，Tick频率或Tick周期是一个最简单的判断指标。

场景2 - 大量同步任务调用

上一种场景，是针对Node应用具有很高的CPU使用率（CPU usage）。而当应用的CPU使用率很低时，却并不意味着Event Loop 中的异步回调没有被大量延迟执行。

一个典型的例子，就是Node代码中存在大量的同步IO操作，比如fs.readFileSync()的调用

总结

阻塞后的Event Loop体现出高延迟低空闲（High event loop lag, low event loop idle）的特点。

典型的阻塞Event Loop的 Node API调用分类成两种情况：

同步的CPU密集型操作

  • 加密操作 - Crypto 模块中crypto.pbkdf2() crypto.randomBytes()`` crypto.randomFill()方法
  • 压缩操作 - Zlib模块中没有显式使用libuv线程池进行同步调用的方法

同步的IO密集型操作

  • DNS模块中的 dns.lookup()和dns.lookupService()
  • File System模块中 除了fs.FSWatcher()和那些没有显式使用libuv线程池进行同步调用的方法

找到event loop的Tick Duration增加的原因，是优化Node应用的一个关键。而将这些同步的API调用放到Node中工作池（Worker Pool）中，是一个有效的解决办法。

场景3 - 内存泄漏

http://pmuellr.github.io/slides/2017/01-profiling-node/demos/express-demo.js.html

工具

profiler inside V8

https://nodejs.org/en/docs/guides/simple-profiling/

Node Application Metrics

Node Application Metrics （https://github.com/RuntimeTools/appmetrics）是一个由IBM开发的专门针对Node.js应用的性能资源监测数据采集工具。

在数据采集完成后，可以通过Node Application Metrics Dashboard（https://github.com/RuntimeTools/appmetrics-dash）将性能指标（Performance Metrics）可视化显示。

Node Application Metrics收集的原始数据包括：

Source Description
Environment Machine and runtime environment information
CPU Process and system CPU
Memory Process and system memory usage
GC Node/V8 garbage collection statistics
Event Loop Event loop latency information
Loop Event loop timing metrics
Function profiling Node/V8 function profiling (disabled by default)
HTTP HTTP request calls made of the application
HTTP Outbound HTTP requests made by the application
socket.io WebSocket data sent and received by the application
LevelDB LevelDB queries made by the application
MySQL MySQL queries made by the application
MongoDB MongoDB queries made by the application
PostgreSQL PostgreSQL queries made by the application
MQTT MQTT messages sent and received by the application
MQLight MQLight messages sent and received by the application
Memcached Data that is stored or manipulated in Memcached
OracleDB OracleDB queries made by the application
Oracle Oracle queries made by the application
StrongOracle StrongOracle database queries made by the application
Redis Redis commands issued by the application
Riak Riak methods called by the application
Request tracking A tree of application requests, events and optionally trace (disabled by default)
Function trace Tracing of application function calls that occur during a request (disabled by default)

Application Metrics for Node.js

npm install -S appmetrics-dash

# In your Node.js application, add:
require('appmetrics-dash').monitor();

# Monitor your running Node.js application in a web browser at:
localhost:3001/appmetrics-dash

easy-monitor

npm install express
npm install easy-monitor
main.js
'use strict';
const easyMonitor = require('easy-monitor');
easyMonitor('Mercury');
const express = require('express');
const app = express();

app.get('/hello', function (req, res, next) {
    res.send('hello');
});

app.listen(8082);

Node Report

https://www.npmjs.com/package/node-report

NodeSource N|Solid Platform

https://nodesource.com/products/nsolid

https://docs.nodesource.com/nsolid/3.4/docs#console-overview

npm module v8-profiler

https://www.youtube.com/watch?v=gL2GGcV_f20

V8 Inspector

https://nodejs.org/dist/latest-v6.x/docs/api/debugger.html#debugger_v8_inspector_integration_for_node_js

  • 分析函数执行时间 - V8 CPU profiler
    • 找到应用中的瓶颈（bottleneck）
    • 优化应用性能
  • 分析内存使用 - V8 heap snapshots profiler
    • 发现内存泄漏
    • 优化程序的内存使用

Reference

https://www.dynatrace.com/news/blog/all-you-need-to-know-to-really-understand-the-node-js-event-loop-and-its-metrics/

https://nodejs.org/en/docs/guides/dont-block-the-event-loop/

https://nodesource.com/blog/node-js-performance-monitoring-part-3-debugging-the-event-loop/

https://blog.ghaiklor.com/profiling-nodejs-applications-1609b77afe4e

https://blog.risingstack.com/monitoring-nodejs-applications-nodejs-at-scale/

https://blog.appdynamics.com/product/top-5-performance-metrics-for-node-js-applications/

https://www.youtube.com/watch?v=9RhOGoChGqo

http://pmuellr.github.io/slides/2017/01-profiling-node/#24

http://pmuellr.github.io/slides/2016/01-intro-to-profiling/intro-to-profiling.pdf

FEATURED TAGS
algorithm algorithmproblem architecturalpattern architecture aws c# cachesystem codis concurrentcontrol database dataformat datastructure debug design designpattern distributedsystem django docker domain engineering freebsd git golang grafana hackintosh hadoop hardware hexo http hugo ios iot java javaee javascript kafka kubernetes linux linuxcommand linuxio lock macos markdown microservices mysql nas network networkprogramming nginx node.js npm oop openwrt operatingsystem padavan performance programming prometheus protobuf python redis router security shell software-testing spring sql systemdesign truenas ubuntu vmware vpn windows wmware wordpress zookeeper
FRIENDS

TOC