0.背景
组内需要一款轻量级的性能测试工具，之前考虑过LR（太笨重，单实例，当然它的地位是不容置疑的），阿里云的PTS( https://pts.aliyun.com/lite/index.htm ， 仅支持阿里云内网和公网机器)，Gatling(http://gatling.io/#/)没有TPS数据等等，不太适合我们。
nGrinderr是NAVER（韩国最大互联网公司NHN旗下搜索引擎网站）开源的性能测试工具，直接部署成web服务，支持多用户使用，可扩展性好，可自定义plugin(http://www.cubrid.org/wiki_ngrinder/entry/how-to-develop-plugin)，wiki文档较丰富(http://www.cubrid.org/wiki_ngrinder/entry/ngrinder-devzone)，数据及图形化展示满足需求；但是展示的统计数据较简单，二次开发整合数据：TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数字段，并将这些数据展示在详细测试报告页中。
1.项目剖析
1-1. nGrinder架构
nGrinder是一款在一系列机器上执行Groovy或JPython测试脚本的应用，内部引擎是基于Grinder。
架构图：
层级图：
默认的NGRINDER_HOME为/root/.ngrinder， 大多是配置文件和数据文件。
目录/root/.ngrinder/perftest/0_999下，以每个test_id为名的文件夹对应的存储了执行性能测试时的采样数据：
*.data文件就是执行性能测试时对应的各种性能采样数据，性能测试详细报告页就是根据这些data文件，进行图形化展示（ajax）。
nGrinder包含2大组件：
1）Controller
为性能测试提供web interface
协同测试进程
收集和显示测试数据
新建和修改脚本
2）Agent
agent mode: 运行进程和线程，压测目标服务
monitor mode: 监控目标系统性能(cpu/memory)， 可以自定义收集的数据(比如 jvm数据)
http://www.cubrid.org/wiki_ngrinder/entry/general-architecture
1-2. 技术栈
1)Controller 层
FreeMarker: 基于Java的模板引擎
Spring Security
Spring Mvc：Spring MVC provides rich functionality for building robust web applications.
GSon
SVNKit Dav
2)Service 层
Grinder
Spring
EhCache: Ehcache has excellent Spring integration.
3)Data层
Spring Data
Hibernate:Hibernate is a powerful technology for persisting data,and it is Spring Data back-end within nGrinder.
H2: (nGrinder默认使用该DB)
Cubrid:(nGrinder同一家公司的DB)
Liquidase: Liquibase is an open source that automates database schema updates.
SVNKit
http://www.cubrid.org/wiki_ngrinder/entry/technology-stack
2.源码实现
需求：在详细测试报告页中展示TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数这些数据。
修改Controller层，增加数据处理业务逻辑(计算TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数)
在获取采样数据
ngrinder-core/src/main/java/net/grinder/SingleConsole.java中新增处理业务逻辑，核心修改代码片段：
// tps list
List
// rt list
List
/**
*
* 每次请求调用一次 Build up statistics for current sampling.
*
* @param accumulatedStatistics
* intervalStatistics
* @param intervalStatistics
* accumulatedStatistics
*/
protected void updateStatistics(StatisticsSet intervalStatistics,
StatisticsSet accumulatedStatistics) {
Map
result.put("testTime", getCurrentRunningTime() / 1000);
List
List
for (Test test : accumulatedStatisticMapPerTest.keySet()) {
Map
Map
StatisticsSet accumulatedSet = this.accumulatedStatisticMapPerTest
.get(test);
StatisticsSet intervalSet = this.intervalStatisticMapPerTest
.get(test);
accumulatedStatisticMap.put("testNumber", test.getNumber());
accumulatedStatisticMap.put("testDescription",
test.getDescription());
intervalStatisticsMap.put("testNumber", test.getNumber());
intervalStatisticsMap.put("testDescription", test.getDescription());
// When only 1 test is running, it's better to use the parametrized
// snapshot.
for (Entry
if (INTERESTING_STATISTICS.contains(each.getKey())) {
accumulatedStatisticMap.put(
each.getKey(),
getRealDoubleValue(each.getValue().getDoubleValue(
accumulatedSet)));
intervalStatisticsMap.put(
each.getKey(),
getRealDoubleValue(each.getValue().getDoubleValue(
intervalSet)));
}
}
cumulativeStatistics.add(accumulatedStatisticMap);
lastSampleStatistics.add(intervalStatisticsMap);
}
Map
for (Entry
if (INTERESTING_STATISTICS.contains(each.getKey())) {
totalStatistics.put(each.getKey(), getRealDoubleValue(each
.getValue().getDoubleValue(accumulatedStatistics)));
}
}
LOGGER.debug("hugang start get plug data");
// 获取tps, rt集合
for (Entry
if ("TPS".equals(each.getKey())) {
tps.add((Double) getRealDoubleValue(each.getValue()
.getDoubleValue(intervalStatistics)));
} else if ("Mean_Test_Time_(ms)".equals(each.getKey())) {
meanTestTime.add((Double) getRealDoubleValue(each.getValue()
.getDoubleValue(intervalStatistics)));
}
}
result.put("totalStatistics", totalStatistics);
result.put("cumulativeStatistics", cumulativeStatistics);
result.put("lastSampleStatistics", lastSampleStatistics);
result.put("tpsChartData", getTpsValues());
result.put("peakTpsForGraph", this.peakTpsForGraph);
synchronized (this) {
result.put(GrinderConstants.P_PROCESS, this.runningProcess);
result.put(GrinderConstants.P_THREAD, this.runningThread);
result.put("success", !isAllTestFinished());
}
// Finally overwrite.. current one.
this.statisticData = result;
}
/**
* 从updateStatistics()累加数据, list :rt 和 tps, 为成员变量
*
* 再处理集合，放到statisticData中
*
* @author hugang
*/
public void getPlusResult(){
LOGGER.debug("hugang getPlusResult() tpslist {} rtlist is {}",
tps.toString(), meanTestTime.toString());
int i = 0;
int j = 0;
// list转成数组， 标准库使用数组作为参数
double[] tpsArray = new double[tps.size()];
for (double tpsNum : tps) {
tpsArray[i++] = tpsNum;
}
// list转成数组
double[] meanTestTimeArray = new double[meanTestTime.size()];
for (double meanTime : meanTestTime) {
meanTestTimeArray[j++] = meanTime;
}
// tps 标准差
double tpsStd = new StandardDeviation().evaluate(tpsArray);
// tps 平均值
double tpsMean = new Mean().evaluate(tpsArray, 0, tpsArray.length);
// tps 波动率= tps 标准差 / tps 平均值
double tpsVix = 0;
if(0 != tpsMean){
tpsVix = tpsStd / tpsMean;
}
// meanTestTime 百分位数
Percentile percentile = new Percentile();
// 先排序
Arrays.sort(meanTestTimeArray);
// meanTestTime最小值
double minMeanTime = meanTestTimeArray[0];
double twentyFiveMeanTime = percentile.evaluate(meanTestTimeArray, 25);
double fiftyMeanTime = percentile.evaluate(meanTestTimeArray, 50);
double serventyFiveMeanTime = percentile
.evaluate(meanTestTimeArray, 75);
double eightyMeanTime = percentile.evaluate(meanTestTimeArray, 80);
double eightyFiveMeanTime = percentile.evaluate(meanTestTimeArray, 85);
double ninetyMeanTime = percentile.evaluate(meanTestTimeArray, 90);
double ninetyFiveMeanTime = percentile.evaluate(meanTestTimeArray, 95);
double ninetyNineMeanTime = percentile.evaluate(meanTestTimeArray, 99);
int length = meanTestTimeArray.length;
// meanTestTime最高值
double maxMeanTime = meanTestTimeArray[length - 1];
// meanTestTime平均值
// double TimeMean = new Mean().evaluate(meanTestTimeArray, 0,
// meanTestTimeArray.length);
LOGGER.debug(
"hugang plug Statistics MinMeanTime {} MaxMeanTime is {}",
minMeanTime, maxMeanTime);
// 附加信息 hugang
// tps 标准差, tps 波动率， 最小/最大RT, RT百分位数
Map
plusStatistics.put("tpsStd", tpsStd);
// plusStatistics.put("tpsMean", tpsMean);
plusStatistics.put("tpsVix", tpsVix);
plusStatistics.put("minMeanTime", minMeanTime);
plusStatistics.put("twentyFiveMeanTime", twentyFiveMeanTime);
plusStatistics.put("fiftyMeanTime", fiftyMeanTime);
plusStatistics.put("serventyFiveMeanTime", serventyFiveMeanTime);
plusStatistics.put("eightyMeanTime", eightyMeanTime);
plusStatistics.put("eightyFiveMeanTime", eightyFiveMeanTime);
plusStatistics.put("ninetyMeanTime", ninetyMeanTime);
plusStatistics.put("ninetyFiveMeanTime", ninetyFiveMeanTime);
plusStatistics.put("ninetyNineMeanTime", ninetyNineMeanTime);
plusStatistics.put("maxMeanTime", maxMeanTime);
LOGGER.debug("SingleConsole plug Statistics map plusStatistics {}", plusStatistics);
this.statisticData.put("plusStatistics", plusStatistics);
}
/**
*
* 停止采样数据
* Stop sampling.
*/
public void unregisterSampling() {
this.currentNotFinishedProcessCount = 0;
if (sampleModel != null) {
this.sampleModel.reset();
this.sampleModel.stop();
}
LOGGER.info("Sampling is stopped");
informTestSamplingEnd();
// 结束采样后，处理数据
// hugang
getPlusResult();
}
Map statisticData为不同数据集集合。
Service层从SingleConsole类中获取数据集statisticData：
ngrinder-controller/src/main/java/org/ngrinder/perftest/server/PerfTestService.java 中Map
/**
* Update the given {@link PerfTest} properties after test finished.
*
* @param perfTest perfTest
*
* getConsoleUsingPort()获取数据
*
*
* hugang
*/
public void updatePerfTestAfterTestFinish(PerfTest perfTest) {
checkNotNull(perfTest);
Map
@SuppressWarnings("unchecked")
Map
// 获取附加数据
Map
LOGGER.info("Total Statistics for test {} is {}", perfTest.getId(), totalStatistics);
LOGGER.info("plug Statistics for test {} is {}", perfTest.getId(), plusStatistics);
perfTest.setTps(parseDoubleWithSafety(totalStatistics, "TPS", 0D));
perfTest.setMeanTestTime(parseDoubleWithSafety(totalStatistics, "Mean_Test_Time_(ms)", 0D));
perfTest.setPeakTps(parseDoubleWithSafety(totalStatistics, "Peak_TPS", 0D));
perfTest.setTests(MapUtils.getDouble(totalStatistics, "Tests", 0D).longValue());
perfTest.setErrors(MapUtils.getDouble(totalStatistics, "Errors", 0D).longValue());
// 附加信息写到model， 持久化
perfTest.setTpsStd(parseDoubleWithSafety(plusStatistics, "tpsStd", 0D));
perfTest.setTpsVix(parseDoubleWithSafety(plusStatistics, "tpsVix", 0D));
perfTest.setMinRT(parseDoubleWithSafety(plusStatistics, "minMeanTime", 0D));
perfTest.setTwentyFiveMeanTime(parseDoubleWithSafety(plusStatistics, "twentyFiveMeanTime", 0D));
perfTest.setFiftyMeanTime(parseDoubleWithSafety(plusStatistics, "fiftyMeanTime", 0D));
perfTest.setServentyFiveMeanTime(parseDoubleWithSafety(plusStatistics, "serventyFiveMeanTime", 0D));
perfTest.setEightyMeanTime(parseDoubleWithSafety(plusStatistics, "eightyMeanTime", 0D));
perfTest.setEightyFiveMeanTime(parseDoubleWithSafety(plusStatistics, "eightyFiveMeanTime", 0D));
perfTest.setNinetyMeanTime(parseDoubleWithSafety(plusStatistics, "ninetyMeanTime", 0D));
perfTest.setNinetyFiveMeanTime(parseDoubleWithSafety(plusStatistics, "ninetyFiveMeanTime", 0D));
perfTest.setNinetyNineMeanTime(parseDoubleWithSafety(plusStatistics, "ninetyNineMeanTime", 0D));
perfTest.setMaxRT(parseDoubleWithSafety(plusStatistics, "maxMeanTime", 0D));
}
修改Model层，在javabean中增加TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数, JPA持久化（H2 DB新增TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数字段)
model文件为：ngrinder-core/src/main/java/org/ngrinder/model/PerfTest.java
/**
* 新增字段，TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数
* hugang
*/
@Expose
@Column(name = "tpsStd")
private Double tpsStd;
@Expose
@Column(name = "tpsVix")
private Double tpsVix;
@Expose
@Column(name = "minRT")
private Double minRT;
@Expose
@Column(name = "twentyFiveMeanTime")
private Double twentyFiveMeanTime;
@Expose
@Column(name = "fiftyMeanTime")
private Double fiftyMeanTime;
@Expose
@Column(name = "serventyFiveMeanTime")
private Double serventyFiveMeanTime;
@Expose
@Column(name = "eightyMeanTime")
private Double eightyMeanTime;
@Expose
@Column(name = "eightyFiveMeanTime")
private Double eightyFiveMeanTime;
@Expose
@Column(name = "ninetyMeanTime")
private Double ninetyMeanTime;
@Expose
@Column(name = "ninetyFiveMeanTime")
private Double ninetyFiveMeanTime;
@Expose
@Column(name = "ninetyNineMeanTime")
private Double ninetyNineMeanTime;
@Expose
@Column(name = "maxRT")
private Double maxRT;
对应的set(), get()
还需修改db change文件(因为系统DB默认使用H2， 只需修改H2对应的xml)，ngrinder-controller/src/main/resources/ngrinder_datachange_logfile/db.changelog_schema_H2.xml
create table PERF_TEST (
id bigint generated by default as identity unique,
created_date timestamp,
last_modified_date timestamp,
agent_count integer,
description varchar(2048),
distribution_path varchar(255),
duration bigint,
errors integer,
finish_time timestamp,
ignore_sample_count integer,
init_processes integer,
init_sleep_time integer,
last_progress_message varchar(2048),
mean_test_time double,
peak_tps double,
errorRate double,
tpsStd double,
tpsVix double,
minRT double,
twentyFiveMeanTime double,
fiftyMeanTime double,
serventyFiveMeanTime double,
eightyMeanTime double,
eightyFiveMeanTime double,
ninetyMeanTime double,
ninetyFiveMeanTime double,
ninetyNineMeanTime double,
maxRT double,
系统重启加载时，Liquidase会自动更新DB。
修改View层，在详细报告对应的freemarker模板新增TPS标准差，TPS波动率，最小/大RT，RT 25/50/75/80/85/90/95/99百分位数字段，前端新增展示这些数据
ngrinder-controller/src/main/webapp/WEB-INF/ftl/perftest/detail_report.ftl
<#-- hugang -->
<#-- 新增 错误率，TPS标准差，TPS波动率，最小RT, 最大RT, RT 25/50/75/80/85/90/95/99百分位数 -->
msmsmsmsmsmsmsmsmsms还有个坑，就是从github拉下的代码，源码中pom.xml依赖的jar包不完整，直接打不了包，项目有的依赖的jar 公有maven仓库已经没有了，需要自己从网上找jar包，安装到本地仓库，我归整了下：
http://download.csdn.net/detail/neven7/9443895
直接在ngrinder根路径下执行打包命令：
mvn -Dmaven.test.skip=true clean package
部署生成的war即可。
3.结果展示
在详细报告页新增如下数据结果：