固定 QPS 压测模式探索
在早前跟测试同行在 QQ 群聊天的时候,聊过一个固定 QPS 压测的问题,最近突然有需求,想实现一下,丰富一下自己的性能测试框架,最新的代码请移步我的GitHub
,地址:https://github.com/JunManYuanLong/FunTester,gitee
地址:https://gitee.com/fanapi/tester。
思路
有一个多线程的基类,其他压测任务类继承于基类。
并发执行类由线程池、任务发生器和补偿器组成。
单线程执行任务发生器将生成的任务对象丢到线程池里面执行。
另起补偿器线程完成缺失的补偿。(由于多种原因,真实发生量小于设定值)
总体的思路与如何mock固定QPS的接口、moco固定QPS接口升级补偿机制这两票文章一致,但是没有采取Semaphore
的模式,原因是moco
是多线程对单线程,压测
是单线程对多线程。
语言继续采用
Java
语言。
基类
写得有点仓促,还未进行大量实践,所以注释少一些。这里依然设计两种子模式:定量压测和定时压测,这里由于两种压测模式,通过一个属性isTimesMode
记录,在执行类FixedQpsConcurrent
中用到,单次压测任务对象统一isTimesMode
和limit
两个属性。
package com.fun.base.constaint;
import com.fun.base.interfaces.MarkThread;
import com.fun.config.HttpClientConstant;
import com.fun.frame.execute.FixedQpsConcurrent;
import com.fun.frame.httpclient.GCThread;
import com.fun.utils.Time;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
public abstract class FixedQpsThread<T> extends ThreadBase {
private static Logger logger = LoggerFactory.getLogger(FixedQpsThread.class);
public int qps;
public int limit;
public boolean isTimesMode;
public FixedQpsThread(T t, int limit, int qps, MarkThread markThread) {
this.limit = limit;
this.qps = qps;
this.mark = markThread;
this.t = t;
isTimesMode = limit > 1000 ? true : false;
}
protected FixedQpsThread() {
super();
}
@Override
public void run() {
try {
before();
threadmark = mark == null ? EMPTY : this.mark.mark(this);
long s = Time.getTimeStamp();
doing();
long e = Time.getTimeStamp();
long diff = e - s;
FixedQpsConcurrent.allTimes.add(diff);
FixedQpsConcurrent.executeTimes.getAndIncrement();
if (diff > HttpClientConstant.MAX_ACCEPT_TIME)
FixedQpsConcurrent.marks.add(diff + CONNECTOR + threadmark);
} catch (Exception e) {
logger.warn("执行任务失败!", e);
logger.warn("执行失败对象的标记:{}", threadmark);
FixedQpsConcurrent.errorTimes.getAndIncrement();
} finally {
after();
}
}
@Override
public void before() {
GCThread.starts();
}
/**
* 子类必需实现改方法,不然调用deepclone方法会报错
*
* @return
*/
public abstract FixedQpsThread clone();
}
执行类
此处补偿线程设计还待优化,中间有两处休眠:一处是循环检测是否需要补偿,一处是单词补偿间隔。尚未提取配置变量,有待后面实践之后进行优化调整。测试结果对象依然采用了原来的,数值和计算方式保持一致,后期也会根据实践结果进行调整,可以关注我的GitHub
及时获取更新。
package com.fun.frame.execute;
import com.fun.base.bean.PerformanceResultBean;
import com.fun.base.constaint.FixedQpsThread;
import com.fun.config.Constant;
import com.fun.frame.Save;
import com.fun.frame.SourceCode;
import com.fun.frame.httpclient.GCThread;
import com.fun.utils.Time;
import com.fun.utils.WriteRead;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.Vector;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import static java.util.stream.Collectors.toList;
/**
* 并发类,用于启动压力脚本
*/
public class FixedQpsConcurrent extends SourceCode {
private static Logger logger = LoggerFactory.getLogger(FixedQpsConcurrent.class);
public static boolean key = false;
public static AtomicInteger executeTimes = new AtomicInteger();
public static AtomicInteger errorTimes = new AtomicInteger();
public static Vector<String> marks = new Vector<>();
/**
* 用于记录所有请求时间
*/
public static Vector<Long> allTimes = new Vector<>();
/**
* 开始时间
*/
public long startTime;
/**
* 结束时间
*/
public long endTime;
public int queueLength;
/**
* 任务描述
*/
public String desc = DEFAULT_STRING;
/**
* 任务集
*/
public List<FixedQpsThread> threads = new ArrayList<>();
/**
* 线程池
*/
ExecutorService executorService;
/**
* @param thread 线程任务
*/
public FixedQpsConcurrent(FixedQpsThread thread) {
this(thread, DEFAULT_STRING);
}
/**
* @param threads 线程组
*/
public FixedQpsConcurrent(List<FixedQpsThread> threads) {
this(threads, DEFAULT_STRING);
}
/**
* @param thread 线程任务
* @param desc 任务描述
*/
public FixedQpsConcurrent(FixedQpsThread thread, String desc) {
this();
this.queueLength = 1;
threads.add(thread);
this.desc = desc + Time.getNow();
}
/**
* @param threads 线程组
* @param desc 任务描述
*/
public FixedQpsConcurrent(List<FixedQpsThread> threads, String desc) {
this();
this.threads = threads;
this.queueLength = threads.size();
this.desc = desc + Time.getNow();
}
private FixedQpsConcurrent() {
executorService = ThreadPoolUtil.createPool(20, 200, 3);
}
/**
* 执行多线程任务
* 默认取list中thread对象,丢入线程池,完成多线程执行,如果没有threadname,name默认采用desc+线程数作为threadname,去除末尾的日期
*/
public PerformanceResultBean start() {
key = false;
FixedQpsThread fixedQpsThread = threads.get(0);
boolean isTimesMode = fixedQpsThread.isTimesMode;
int limit = fixedQpsThread.limit;
int qps = fixedQpsThread.qps;
long interval = 1_000_000_000 / qps;
AidThread aidThread = new AidThread();
new Thread(aidThread).start();
startTime = Time.getTimeStamp();
while (true) {
executorService.execute(threads.get(limit-- % queueLength).clone());
if (key ? true : isTimesMode ? limit < 1 : Time.getTimeStamp() - startTime > fixedQpsThread.limit) break;
sleep(interval);
}
endTime = Time.getTimeStamp();
aidThread.stop();
GCThread.stop();
try {
executorService.shutdown();
executorService.awaitTermination(10, TimeUnit.SECONDS);
} catch (InterruptedException e) {
logger.error("线程池等待任务结束失败!", e);
}
logger.info("总计执行 {} ,共用时:{} s,执行总数:{},错误数:{}!", fixedQpsThread.isTimesMode ? fixedQpsThread.limit + "次任务" : "秒", Time.getTimeDiffer(startTime, endTime), executeTimes, errorTimes);
return over();
}
private PerformanceResultBean over() {
key = true;
Save.saveLongList(allTimes, "data/" + queueLength + desc);
Save.saveStringListSync(marks, MARK_Path.replace(LONG_Path, EMPTY) + desc);
allTimes = new Vector<>();
marks = new Vector<>();
executeTimes.set(0);
errorTimes.set(0);
return countQPS(queueLength, desc, Time.getTimeByTimestamp(startTime), Time.getTimeByTimestamp(endTime));
}
/**
* 计算结果
* <p>此结果仅供参考</p>
*
* @param name 线程数
*/
public PerformanceResultBean countQPS(int name, String desc, String start, String end) {
List<String> strings = WriteRead.readTxtFileByLine(Constant.DATA_Path + name + desc);
int size = strings.size();
List<Integer> data = strings.stream().map(x -> changeStringToInt(x)).collect(toList());
int sum = data.stream().mapToInt(x -> x).sum();
Collections.sort(data);
String statistics = StatisticsUtil.statistics(data, desc, this.queueLength);
double qps = 1000.0 * size * name / sum;
return new PerformanceResultBean(desc, start, end, name, size, sum / size, qps, getPercent(executeTimes.get(), errorTimes.get()), 0, executeTimes.get(), statistics);
}
/**
* 用于做后期的计算
*
* @param name
* @param desc
* @return
*/
public PerformanceResultBean countQPS(int name, String desc) {
return countQPS(name, desc, Time.getDate(), Time.getDate());
}
/**
* 后期计算用
*
* @param name
* @return
*/
public PerformanceResultBean countQPS(int name) {
return countQPS(name, EMPTY, Time.getDate(), Time.getDate());
}
/**
* 补偿线程
*/
class AidThread implements Runnable {
private boolean key = true;
int i;
public AidThread() {
}
@Override
public void run() {
logger.info("补偿线程开始!");
while (key) {
long expect = (Time.getTimeStamp() - startTime) / 1000 * threads.get(0).qps;
if (expect > executeTimes.get() + 10) {
range((int) expect - executeTimes.get()).forEach(x -> {
sleep(100);
executorService.execute(threads.get(i++ % queueLength).clone());
});
}
sleep(3);
}
logger.info("补偿线程结束!");
}
public void stop() {
key = false;
}
}
}
其他配套的标记类
、统计类
还等待修改,比较简单,这里不放代码了。
FunTester 热文精选
版权声明: 本文为 InfoQ 作者【FunTester】的原创文章。
原文链接:【http://xie.infoq.cn/article/ae51ad8fe1aba21d4469a1f67】。文章转载请联系作者。
FunTester
公众号:FunTester,650+原创,欢迎关注 2020.10.20 加入
Have Fun,Tester! 公众号FunTester,坚持原创文章的测试人。 FunTester测试框架作者,DCS_FunTester分布式性能测试框架作者。
评论