elasticsearch性能测试工具rally深入详解

题记

elasticsearch性能测试研究了很久，自己想过通过批量导入数据，然后记录时间，统计CPU、内存等变化，计算得出某个性能指标。但显然，数据量起伏不定，非常不准确。
研究发现，github上提供了rally作为elasticsearch的性能测试工具，较好的解决了es性能测试问题。

1、esrally功能：es的性能测试工具。

esrally不支持windows版本，目前只支持Linux和Mac OS。
esrally的运行基于以下两点假设：
1）所有运行在同一台机器完成。(未来改假设条件可能会去掉）
2）你需要向Elasticsearch索引中添加特定的数据集，然后在其上进行基准查询（benchmarking queries）.

2、esrally安装：

前提:验证是否安装了以下内容（注意版本）
Python 3.4+ available as python3 on the path (verify with: python3 –version which should print Python 3.4.0 or higher)
pip3 available on the path (verify with pip3 –version)
JDK 8
git

如下所示：

[root@laoyang git-2.4.0]# pythonPython 3.5.1 (default, Aug 2 2016, 09:53:20)[GCC 4.4.7 20120313 (Red Hat 4.4.7-17)] on linuxType "help", "copyright", "credits" or "license" for more information.[root@laoyang git-2.4.0]# git --versiongit version 2.4.0[root@laoyang git-2.4.0]# java -versionjava version "1.8.0_91"Java(TM) SE Runtime Environment (build 1.8.0_91-b14)Java HotSpot(TM) 64-Bit Server VM (build 25.91-b14, mixed mode)[root@laoyang git-2.4.0]# find / -name "pip3"/usr/bin/pip3/usr/local/python3/bin/pip3

步骤1：安装 Rally

pip3 install esrally

步骤2：配置 Rally

调用： esrally configure 会提示让你输入 java8.home（也就是JAVA_HOME设定的环境变量路径）。
仅java8.home使用自己本机的JAVA_HOME环境变量路径，其他采用默认值即可。

[root@laoyang .rally]# cat /root/.rally/rally.ini[meta]config.version = 5[system]root.dir = /root/.rally/benchmarkslog.root.dir = logsenv.name = local[source]local.src.dir=/root/.rally/src   #注意，此路径并非elaticsearch安装路径。remote.repo.url=https://github.com/elastic/elasticsearch.git[provisioning]local.install.dir = install[runtime]java8.home = /opt/jdk1.8.0_91  #注意，JAVA_HOME 路径为：/opt/jdk1.8.0_91[benchmarks]local.dataset.cache = ${system:root.dir}/data[reporting]datastore.type = elaticsearchdatastore.host = 10.0.1.30datastore.port = 9200datastore.secure = Falsedatastore.user =datastore.password =[tracks]default.url = https://github.com/elastic/rally-tracks

步骤3：运行esrally

由于elasticsearch的运行必须非root账户。esrally建议用非root账户执行。

3、esrally执行常见问题及解决方案

问题1：版本用哪个版本？

目前用最新版本 Rally 0.3.2（2016-8-7发行）。

问题2：注意gradle需要2.13版本，高版本会不兼容。

[root@laoyang caches]# ll /usr/bin/gradle
lrwxrwxrwx 1 root root 27 Aug 5 17:13 /usr/bin/gradle -> /opt/gradle-2.13/bin/gradle

问题3：安装时候提示 documents.json.bz2 无法下载成功。

1）手动下载： http://benchmarks.elasticsearch.org.s3.amazonaws.com/corpora/geonames/documents.json.bz2

2）移动到 /root/.rally/benchmarks/data/geonames/（对于root用户）

对于非root用户,elaticsearch用户，移动路径为：/home/elasticsearch/benchmarks/data/geonames/

问题4: esrally单独运行和加参数运行区别？

esrally单独运行：非常方便的指令，等价于 :esrally –pipeline=from-sources –version=current. Rally 将要通过Gradle从源码编译elasticsearch。
esrally –pipline=from-distribution –distribution-version=2.3.4, Rally将要从 https://download.elasticsearch.org/elasticsearch/release/org/elasticsearch/distribution/tar/elasticsearch/2.3.4/elasticsearch-2.3.4.tar.gz

下载elaticsearch2.3.4版本运行。

问题5： Could not resolve all dependencies for configuration ‘:benchmarks:classpath’.

37 > Could not download shadow.jar (com.github.jengelman.gradle.plugins:shadow:1.2.3)如何解决？
root账户下：

cd /home/elasticsearch/.rally/benchmarks/src
gradle :distribution:tar:assemble

正确提示信息为：

deprecated API.Note: Recompile with -Xlint:deprecation for details.:modules:transport-netty4:processResources UP-TO-DATE:modules:transport-netty4:classes:modules:transport-netty4:jar:modules:transport-netty4:copyPluginPropertiesTemplate:modules:transport-netty4:pluginProperties:modules:transport-netty4:bundlePlugin:distribution:buildModules:distribution:tar:buildTar:distribution:tar:generatePomFileForNebulaPublication:distribution:tar:signArchives SKIPPED:distribution:tar:assembleBUILD SUCCESSFUL

4、esrally最终运行结果：

Preparing for race (might take a few moments) ...  Building from sources .../home/elasticsearch/.rally/benchmarks/races/2016-08-08-07-29-59/local/logs/rally_out.logRacing on track [geonames] and challenge [append-no-conflicts] with car [defaults]  Benchmarking indexing at 12183.1 docs/s [100% done]  Benchmarking stats (warmup iteration 100/100) [100% done]  Benchmarking stats (iteration 100/100) [100% done]  Benchmarking search (warmup iteration 100/100) [100% done]  Benchmarking search (iteration 100/100) [100% done]------------------------------------------------------  _______ __ _____  / ____(_)___ ____ _/ / / ___/_________ ________  / /_ / / __ \/ __ `/ / \__ \/ ___/ __ \/ ___/ _ \ / __/ / / / / / /_/ / / ___/ / /__/ /_/ / / / __//_/ /_/_/ /_/\__,_/_/ /____/\___/\____/_/ \___/------------------------------------------------------| Metric | Value ||----------------------------------------------------------:|----------:|| Min Indexing Throughput [docs/s] | 11949 || Median Indexing Throughput [docs/s] | 12307 || Max Indexing Throughput [docs/s] | 13349 || Indexing time [min] | 164.045 || Merge time [min] | 32.3815 || Refresh time [min] | 8.82333 || Flush time [min] | 1.63852 || Merge throttle time [min] | 1.45482 || Query latency default (90.0 percentile) [ms] | 68.8676 || Query latency default (99.0 percentile) [ms] | 77.6009 || Query latency default (100 percentile) [ms] | 78.8328 || Query latency term (90.0 percentile) [ms] | 4.63227 || Query latency term (99.0 percentile) [ms] | 10.9506 || Query latency term (100 percentile) [ms] | 13.0573 || Query latency phrase (90.0 percentile) [ms] | 6.35018 || Query latency phrase (99.0 percentile) [ms] | 13.1745 || Query latency phrase (100 percentile) [ms] | 14.2399 || Query latency country_agg_uncached (90.0 percentile) [ms] | 361.913 || Query latency country_agg_uncached (99.0 percentile) [ms] | 384.75 || Query latency country_agg_uncached (100 percentile) [ms] | 392.645 || Query latency country_agg_cached (90.0 percentile) [ms] | 7.71451 || Query latency country_agg_cached (99.0 percentile) [ms] | 17.4704 || Query latency country_agg_cached (100 percentile) [ms] | 18.4851 || Query latency scroll (90.0 percentile) [ms] | 50.4234 || Query latency scroll (99.0 percentile) [ms] | 50.9866 || Query latency scroll (100 percentile) [ms] | 51.2115 || Query latency expression (90.0 percentile) [ms] | 482.02 || Query latency expression (99.0 percentile) [ms] | 485.951 || Query latency expression (100 percentile) [ms] | 491.999 || Query latency painless_static (90.0 percentile) [ms] | 890.382 || Query latency painless_static (99.0 percentile) [ms] | 918.354 || Query latency painless_static (100 percentile) [ms] | 927.282 || Query latency painless_dynamic (90.0 percentile) [ms] | 968.908 || Query latency painless_dynamic (99.0 percentile) [ms] | 1019.14 || Query latency painless_dynamic (100 percentile) [ms] | 1021.97 || Median CPU usage (index) [%] | 887.7 || Median CPU usage (stats) [%] | 94.9 || Median CPU usage (search) [%] | 445.05 || Total Young Gen GC [s] | 89.121 || Total Old Gen GC [s] | 12.274 || Index size [GB] | 3.30111 || Totally written [GB] | 20.2123 || Heap used for segments [MB] | 21.6794 || Heap used for doc values [MB] | 0.113129 || Heap used for terms [MB] | 20.0574 || Heap used for norms [MB] | 0.0761719 || Heap used for points [MB] | 0.770404 || Heap used for stored fields [MB] | 0.6623 || Segment count | 96 || Indices Stats(90.0 percentile) [ms] | 11.5116 || Indices Stats(99.0 percentile) [ms] | 14.1724 || Indices Stats(100 percentile) [ms] | 36.8348 || Nodes Stats(90.0 percentile) [ms] | 11.342 || Nodes Stats(99.0 percentile) [ms] | 13.435 || Nodes Stats(100 percentile) [ms] | 16.4768 |

Logs for this race are archived in /home/elasticsearch/.rally/benchmarks/races/2016-08-08-07-29-59/local/logs-geonames-append-no-conflicts-defaults.zip

5、问题大讨论

https://discuss.elastic.co/t/the-below-bugs-appers-when-running-esrally/57063/25

源码 作者danielmitterdorfer一一回复， 最终作者回复：
I am very happy too and I am glad you had so much persistence. You also uncovered a few usability problems that I need to tackle.
帮作者提了一些建议，促使他升级了一个版本0.3.2。

参考：https://github.com/elastic/rally

小结

前面的确费劲周折，花费了接近3整天实践，期间甚至告诉原作者工具太难用，给作者反复沟通，最终问题解决。看到输出结果的那一刻，整个人很兴奋。坚持到底，相信付出的力量！不断思考、反思中前行！

2016-08-08 21:00思于家中窗前
作者：铭毅天下
转载请标明出处，原文地址：
http://blog.csdn.net/laoyang360/article/details/52155481
如果感觉本文对您有帮助，请点击‘顶’支持一下，您的支持是我坚持写作最大的动力，谢谢！