nGrinder源码分析:自动中断测试任务
2016-03-13 10:09
260 查看
1.背景
在运行nGrinder任务时,会出现任务被系统中断,为什么会自动中断呢?是bug还是一种保护机制?本文通过源码分析的方式来解读。2.源码分析
之前的一篇文章:性能测试工具 nGrinder 项目剖析及二次开发,介绍了nGrinder的整体架构,知道ngrinder-core/src/main/java/net/grinder/SingleConsole.java是将测试脚本发布到Agent并在其上执行性能测试,该类实现了收集测试过程中的采样数据,将数据分成两部分记录:写入记录文件,供时序图展示(*.data和csv)
写入内存,供详细数据展示(
private Map<String, Object> statisticData, 持久化到DB)
其中将采样数据写到第一部分(*.data和csv)的实现方法如下:
/* * (non-Javadoc) * * @see * net.grinder.console.model.SampleListener#update(net.grinder.statistics * .StatisticsSet, net.grinder.statistics.StatisticsSet) */ @Override public void update(final StatisticsSet intervalStatistics, final StatisticsSet cumulativeStatistics) { try { if (!capture) { return; } samplingCount++; long currentPeriod = cumulativeStatistics.getValue(getSampleModel() .getPeriodIndex()); setTpsValue(sampleModel.getTPSExpression().getDoubleValue( intervalStatistics)); checkTooLowTps(getTpsValues()); updateStatistics(intervalStatistics, cumulativeStatistics); // 将采样数据写入csv数据 // hugang writeIntervalCsvData(intervalStatistics); int interval = getSampleModel().getSampleInterval(); long gap = 1; if (samplingCount == 1) { lastSamplingPeriod = currentPeriod; } else { lastSamplingPeriod = lastSamplingPeriod + interval; gap = ((currentPeriod - lastSamplingPeriod) / interval); } // Adjust sampling delay.. run write data multiple times... when it // takes longer than 1 // sec. samplingLifeCycleListener .apply(new Informer<SamplingLifeCycleListener>() { @Override public void inform(SamplingLifeCycleListener listener) { listener.onSampling(getReportPath(), intervalStatistics, cumulativeStatistics); } }); for (int i = 0; i < (gap + 1); i++) { final boolean lastCall = (samplingCount == 1 && i == 0) || (samplingCount != 1 && i == gap); // 将采样数据写到*.data文件 // hugang writeIntervalSummaryData(intervalStatistics, lastCall); if (interval >= (MIN_SAMPLING_INTERVAL_TO_ACTIVATE_TPS_PER_TEST)) { writeIntervalSummaryDataPerTest( intervalStatisticMapPerTest, lastCall); } samplingLifeCycleFollowupListener .apply(new Informer<SamplingLifeCycleFollowUpListener>() { @Override public void inform( SamplingLifeCycleFollowUpListener listener) { listener.onSampling(getReportPath(), intervalStatistics, cumulativeStatistics, lastCall); } }); } checkTooManyError(cumulativeStatistics); lastSamplingPeriod = lastSamplingPeriod + (interval * gap); } catch (RuntimeException e) { LOGGER.error("Error occurred while updating the statistics : {}", e.getMessage()); LOGGER.debug("Details : ", e); throw e; } }
其中:
writeIntervalCsvData(intervalStatistics); writeIntervalSummaryData(intervalStatistics, lastCall);
分别将采样数据写到output.csv和*.data。
注意,有2个check方法:
checkTooLowTps(getTpsValues()); checkTooManyError(cumulativeStatistics);
checkTooLowTps(getTpsValues());会判断1分钟内TPS小于0.01,如果为true,将向ConsoleShutdownListener监听器发送停止信号。
/** * Check if the TPS is too low. the TPS is lower than 0.001 for 1 minutes, * It emits a shutdown event to the {@link ConsoleShutdownListener} * * @param tps * current TPS */ private void checkTooLowTps(double tps) { // If the tps is too low, which means the agents or scripts went wrong. if (tps < 0.001) { if (momentWhenTpsBeganToHaveVerySmall == 0) { momentWhenTpsBeganToHaveVerySmall = System.currentTimeMillis(); } else if (new Date().getTime() - momentWhenTpsBeganToHaveVerySmall >= TOO_LOW_TPS_TIME) { LOGGER.warn( "Stop the test because its tps is less than 0.001 for more than {} minitue.", TOO_LOW_TPS_TIME / 60000); getListeners().apply(new Informer<ConsoleShutdownListener>() { public void inform(ConsoleShutdownListener listener) { listener.readyToStop(StopReason.TOO_LOW_TPS); } }); momentWhenTpsBeganToHaveVerySmall = 0; } } else { momentWhenTpsBeganToHaveVerySmall = 0; } }
private void checkTooManyError(StatisticsSet cumulativeStatistics):会判断10s内事务数错误率>=50%,如果为true,通知监听器ConsoleShutdownListener listener终止任务。
/** * Check if too many error has been occurred. If the half of total * transaction is error for the last 10 secs. It notifies the * {@link ConsoleShutdownListener} * * @param cumulativeStatistics * accumulated Statistics */ private void checkTooManyError(StatisticsSet cumulativeStatistics) { StatisticsIndexMap statisticsIndexMap = getStatisticsIndexMap(); long testSum = cumulativeStatistics.getCount(statisticsIndexMap .getLongSampleIndex("timedTests")); long errors = cumulativeStatistics.getValue(statisticsIndexMap .getLongIndex("errors")); // testSum 成功事务数, errors 失败事务数 // hugang if (((double) (testSum + errors)) / 2 < errors) { if (lastMomentWhenErrorsMoreThanHalfOfTotalTPSValue == 0) { lastMomentWhenErrorsMoreThanHalfOfTotalTPSValue = System .currentTimeMillis(); } else if (isOverLowTpsThreshold()) { LOGGER.warn( "Stop the test because the count of test error is more than" + " half of total tps for last {} seconds.", TOO_MANY_ERROR_TIME / 1000); getListeners().apply(new Informer<ConsoleShutdownListener>() { public void inform(ConsoleShutdownListener listener) { listener.readyToStop(StopReason.TOO_MANY_ERRORS); } }); lastMomentWhenErrorsMoreThanHalfOfTotalTPSValue = 0; } } }
3.总结
nGrinder中断测试任务,是一种保护机制,当被测系统性能已经很差,nGrinder不会继续对该系统产生压力。判断标准:
TPS在1分钟内小于0.001事务错误率在10s内大于等于50%
如果上述条件,某一条为true,自动中断任务。
附:
nGrinder的debug日志为:
/root/.ngrinder/logs/ngrinder.log
相关文章推荐
- HDU3785寻找大富翁~~真真切切的水题
- 记忆体
- C语言学习——单链表
- 持久化API(JPA)系列(三)实体Bean的开发技术-建立与数据库的连接
- Java虚拟机内存区域堆(heap)的管理
- java异常处理,GUI图形用户界面总结
- 机房重构—单例模式
- Xcode快捷键
- Android Broadcast详解
- CodeForces 86D Powerful array 莫队算法
- 前m大的数
- 剑指offer 合并两个排序的链表
- 使用easy_install安装numpy、pandas、matplotlib及各种第三方模块
- [兴趣] 改造公交卡
- python 输出大文本文件
- 新增列并添加注释
- Java 开发设计原则
- 各种不会各种学
- python selenium2 环境搭建
- 《数据结构与算法分析C++描述(第3版)》第三章List容器可编译运行版