安装

Python 3.8+ including pip3, git 1.9+

pip3 install esrally

高级配置

esrally configure --advanced-config （可多次执行）

Enter the benchmark root directory (default: /home/apps/.rally/benchmarks): 
  Using default value '/home/apps/.rally/benchmarks'

Enter your Elasticsearch project directory: (default: /home/apps/.rally/benchmarks/src/elasticsearch): 
  Using default value '/home/apps/.rally/benchmarks/src/elasticsearch'

Where should metrics be kept?

(1) In memory (simpler but less options for analysis)
(2) Elasticsearch (requires a separate ES instance, keeps all raw samples for analysis)

 (default: 1): 2

Enter the host name of the ES metrics store (default: localhost): 
  Using default value 'localhost'

Enter the port of the ES metrics store: 9200

Use secure connection (True, False) (default: False): 

Username for basic authentication (empty if not needed) (default: ): 

Password for basic authentication (empty if not needed) (default: ): 

Enter a descriptive name for this benchmark environment (ASCII, no spaces) (default: local): rally_demo

Do you want Rally to keep the Elasticsearch benchmark candidate installation including the index (will use several GB per trial run)? (default: False): 

Configuration successfully written to /home/apps/.rally/rally.ini. Happy benchmarking!

• rally-metrics-*：记录race中所有的采样数据（性能报告数据都是基于采样数据汇总得出的结果），可以通过采样数据看到具体抖动情况进行具体分析
• rally-results-*：基于上面的采样数据进行详细的数据汇总，针对每个指标进行汇总，可以针对每个指标进行最终的结果展示
• rally-races-：把最终性能压测结果（即我们看到的输出数据）展示出来，一个文档就是一次race压测

将测试结果放入开启x_pack集群中需要配置

vim .rally/rally.ini

[reporting]

#datastore.ssl.verification_mode = none

datastore.ssl.certificate_authorities = /home/elasticsearch/client-ca.cer

使用

配置代理下载track

export http_proxy=http://proxy.acme.org:8888/

/root/.rally/logs/rally.log将记录

Rally connects via proxy URL [http://proxy.acme.org:3128/] to the Internet (picked up from the environment variable [http_proxy]).

esrally list tracks

类似于赛道的意思

nested 嵌套文档         
percolator 过滤查询     
eventdata              
geoshape               
geopoint 地理查询       
so
geopointshape          
noaa                   
metricbeat             
nyc_taxis 高度数据化结构
http_logs              
pmc 全文搜索            
geonames 结构化数据     

对现有集群进行基准测试

esrally --track=pmc --target-hosts=10.5.5.10:9243,10.5.5.11:9243,10.5.5.12:9243 --pipeline=benchmark-only --client-options="use_ssl:true,verify_certs:true,basic_auth_user:'elastic',basic_auth_password:'changeme'" --user-tag="intention:baseline_github_1234" --report-file=result.csv --report-format=csv

rally可以组成分布式集群，并可以根据配置文件进行测试，因为没有试过，这里不再赘述

esrally list races 对比两场比赛

esrally compare --baseline=0bfd4542-3821-4c79-81a2-0858636068ce --contender=beb154e4-0a05-4f45-ad9f-e34f9a9e51f7

创建track

mkdir .rally/benchmarks/data/XXX

1. 获取数据 allCountries.txt
2. 转换为json格式

vim toJSON.py
import json

cols = (("geonameid", "int", True),
        ("name", "string", True),
        ("asciiname", "string", False),
        ("alternatenames", "string", False),
        ("latitude", "double", True),
        ("longitude", "double", True),
        ("feature_class", "string", False),
        ("feature_code", "string", False),
        ("country_code", "string", True),
        ("cc2", "string", False),
        ("admin1_code", "string", False),
        ("admin2_code", "string", False),
        ("admin3_code", "string", False),
        ("admin4_code", "string", False),
        ("population", "long", True),
        ("elevation", "int", False),
        ("dem", "string", False),
        ("timezone", "string", False))


def main():
    with open("allCountries.txt", "rt", encoding="UTF-8") as f:
        for line in f:
            tup = line.strip().split("\t")
            record = {}
            for i in range(len(cols)):
                name, type, include = cols[i]
                if tup[i] != "" and include:
                    if type in ("int", "long"):
                        record[name] = int(tup[i])
                    elif type == "double":
                        record[name] = float(tup[i])
                    elif type == "string":
                        record[name] = tup[i]
            print(json.dumps(record, ensure_ascii=False))


if __name__ == "__main__":
    main()

python3 toJSON.py > documents.json

3. 编写track.json

4. 检查是否可用esrally info --track-path=~/rally-tracks/tutorial

track解剖

{% set index_count = 2 %} #设置循环次数
{
  "version": 2,
  "description": "Tutorial benchmark for Rally",
  "indices": [
    {
      "name": "geonames", #要操作索引的名称
      "body": "index.json"
    }
  ],
  "corpora": [ #此轨道使用的所有文档语料库
    {
      "name": "rally-tutorial",
      "documents": [
        {
          "source-file": "documents.json",
          "document-count": 11658903,
          "uncompressed-bytes": 1544799789
        }
      ]
    }
  ],
  "schedule": [ #具体执行的操作
    {
      "operation": {
        "operation-type": "delete-index" #操作类型
      }
    },
    {
      "operation": {
        "operation-type": "create-index"
      }
    },
    {
      "operation": {
        "operation-type": "cluster-health",
        "request-params": {
          "wait_for_status": "green"
        }
      }
    },
    {
      "operation": {
        "operation-type": "bulk",
        "bulk-size": 5000
      },
      "warmup-time-period": 120,
      "clients": 8
    },
    {
      "operation": {
        "operation-type": "force-merge"
      }
    },
    {
      "operation": {
        "name": "query-match-all",
        "operation-type": "search",
        "body": {
          "query": {
            "match_all": {}
          }
        }
      },
      "clients": 8, #发起请求客户端数量
      "warmup-iterations": 1000, #预热，不显示在测量结果中
      "iterations": 1000, #执行次数
      "target-throughput": 100 #尽可能慢地运行任务，target-throughput/clients/s qps
    }
  ]
}

配置parallel实现并行执行语句

https://esrally.readthedocs.io/en/latest/track.html

https://elasticsearch-benchmarks.elastic.co/#

https://learnku.com/articles/41047