当前位置： 首页 > news >正文

随笔之TDengine基准测试示例

news 2025/7/31 9:15:46

文章目录

    • 一、基本信息
    • 二、基准测试策略
    • 三、基准测试过程
      • 1. 模拟高并发写入场景
      • 2. 模拟并发查询场景
    • 四、基准测试结论

一、基本信息

  • TDengine 版本：3.3.6.13（目前最新版本）
  • 服务器配置：16核CPU，32GB内存，高IO 1000GB存

二、基准测试策略

  • 测试数据库：
    • 数据库：test
    • 超级表：meters（含字段 ts, current, voltage, phase 和标签 groupid, location）
  • 测试工具：taosBenchmark（TDengine 自带的性能测试工具）
  • 测试目标​​：
    • 模拟高并发写入场景，评估数据库的写入性能。
    • 模拟并发查询场景，评估数据库的查询性能。

三、基准测试过程

1. 模拟高并发写入场景

root@sjfwq-v1-p-0-107:~# taosBenchmark -y
[07/21 11:48:25.593007] INFO: client version: 3.3.6.13Connect mode is : Native[07/21 11:48:26.910499] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.188987] INFO: command to create database: <CREATE DATABASE IF NOT EXISTS test PRECISION 'ms';>
[07/21 11:48:27.189020] SUCC: created database (test)
[07/21 11:48:27.189332] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.191098] INFO: stable meters does not exist, will create one
[07/21 11:48:27.191125] INFO: create stable: <CREATE TABLE IF NOT EXISTS test.meters (ts TIMESTAMP,current float,voltage int,phase float) TAGS (groupid int,location binary(24))>
[07/21 11:48:27.191389] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.200606] INFO: generate stable<meters> columns data with lenOfCols<80> * prepared_rand<20000>
[07/21 11:48:27.219827] INFO: restful connect -> convertServAddr host=localhost port:6041 to serv_addr=0x2ad77f24 iface=0 
[07/21 11:48:27.219839] INFO: start creating 10000 table(s) with 8 thread(s)
[07/21 11:48:27.220192] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.220668] INFO: thread[0] start creating table from 0 to 1249
[07/21 11:48:27.220788] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.221215] INFO: thread[1] start creating table from 1250 to 2499
[07/21 11:48:27.221367] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.221881] INFO: thread[2] start creating table from 2500 to 3749
[07/21 11:48:27.221922] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.222315] INFO: thread[3] start creating table from 3750 to 4999
[07/21 11:48:27.222393] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.223102] INFO: thread[4] start creating table from 5000 to 6249
[07/21 11:48:27.223197] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.223581] INFO: thread[5] start creating table from 6250 to 7499
[07/21 11:48:27.223683] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.224055] INFO: thread[6] start creating table from 7500 to 8749
[07/21 11:48:27.224163] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.225133] INFO: thread[7] start creating table from 8750 to 9999
[07/21 11:48:27.662916] SUCC: Spent 0.4430 seconds to create 10000 table(s) with 8 thread(s) speed: 22573 tables/s, already exist 0 table(s), actual 10000 table(s) pre created, 0 table(s) will be auto created
[07/21 11:48:27.663173] INFO: init pthread_join 0 ...
[07/21 11:48:27.663240] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663282] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663320] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663243] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663362] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663362] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663409] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663418] SUCC: host:(null) port:0  connect successfully.
[07/21 11:48:27.663853] INFO: init pthread_join 1 ...
[07/21 11:48:27.663864] INFO: init pthread_join 2 ...
[07/21 11:48:27.663949] INFO: init pthread_join 3 ...
[07/21 11:48:27.663956] INFO: init pthread_join 4 ...
[07/21 11:48:27.663983] INFO: init pthread_join 5 ...
[07/21 11:48:27.664106] INFO: init pthread_join 6 ...
[07/21 11:48:27.664125] INFO: init pthread_join 7 ...
[07/21 11:48:27.664188] INFO: Estimate memory usage: 12.20MB
[07/21 11:48:27.664196] INFO: run insert thread. real nthread=8
[07/21 11:48:27.664234] INFO: thread[0] start progressive inserting into table from 0 to 1250
[07/21 11:48:27.664255] INFO: thread[1] start progressive inserting into table from 1250 to 2500
[07/21 11:48:27.664278] INFO: thread[2] start progressive inserting into table from 2500 to 3750
[07/21 11:48:27.664295] INFO: thread[3] start progressive inserting into table from 3750 to 5000
[07/21 11:48:27.664323] INFO: thread[4] start progressive inserting into table from 5000 to 6250
[07/21 11:48:27.664358] INFO: thread[5] start progressive inserting into table from 6250 to 7500
[07/21 11:48:27.664381] INFO: thread[6] start progressive inserting into table from 7500 to 8750
[07/21 11:48:27.664388] INFO: pthread_join 0 ...
[07/21 11:48:27.664401] INFO: thread[7] start progressive inserting into table from 8750 to 10000
[07/21 11:48:52.467670] SUCC: thread[1] progressive mode, completed total inserted rows: 12500000, 537815.02 records/second
[07/21 11:48:52.528920] SUCC: thread[3] progressive mode, completed total inserted rows: 12500000, 535889.46 records/second
[07/21 11:48:52.644246] SUCC: thread[7] progressive mode, completed total inserted rows: 12500000, 534693.89 records/second
[07/21 11:48:52.693441] SUCC: thread[4] progressive mode, completed total inserted rows: 12500000, 533009.17 records/second
[07/21 11:48:52.753257] SUCC: thread[0] progressive mode, completed total inserted rows: 12500000, 532702.66 records/second
[07/21 11:48:52.753402] INFO: pthread_join 1 ...
[07/21 11:48:52.753416] INFO: pthread_join 2 ...
[07/21 11:48:52.772948] SUCC: thread[5] progressive mode, completed total inserted rows: 12500000, 532017.61 records/second
[07/21 11:48:52.879224] SUCC: thread[6] progressive mode, completed total inserted rows: 12500000, 529537.68 records/second
[07/21 11:48:53.156213] SUCC: thread[2] progressive mode, completed total inserted rows: 12500000, 524104.91 records/second
[07/21 11:48:53.156314] INFO: pthread_join 3 ...
[07/21 11:48:53.156329] INFO: pthread_join 4 ...
[07/21 11:48:53.156374] INFO: pthread_join 5 ...
[07/21 11:48:53.156401] INFO: pthread_join 6 ...
[07/21 11:48:53.156413] INFO: pthread_join 7 ...
[07/21 11:48:53.157329] SUCC: Spent 25.492230 (real 23.476738) seconds to insert rows: 100000000 with 8 thread(s) into test 3922763.92 (real 4259535.55) records/second
[07/21 11:48:53.157361] SUCC: insert delay, min: 8.7650ms, avg: 18.7814ms, p90: 17.1450ms, p95: 29.5320ms, p99: 366.1130ms, max: 670.7390ms
[07/21 11:48:53.157378] INFO: free resource and exit ...

结果解读：

  • 数据库初始化与表创建测试​：创建超级表、子表创建 (10,000张)，耗时：​​0.443秒​​，​​22,573表/秒​​
  • 数据写入测试​：数据量（1亿行），写入线程（8线程），总耗时（​​25.49秒​​），写入吞吐量（​​3,922,764行/秒​​），平均延迟（18.78ms）

2. 模拟并发查询场景

root@sjfwq-v1-p-0-107:~/taosBenchmark# taosBenchmark -f query.json 
[07/21 17:02:22.358884] INFO: query.json
{"filetype":	"query","cfgdir":	"/etc/taos","host":	"127.0.0.1","port":	6030,"user":	"root","password":	"taosdata","confirm_parameter_prompt":	"no","continue_if_fail":	"yes","databases":	"test","query_times":	10,"query_mode":	"taosc","specified_table_query":	{"query_interval":	1,"threads":	3,"sqls":	[{"sql":	"select last_row(*) from meters","result":	"./query_res0.txt"}, {"sql":	"select count(*) from d0","result":	"./query_res1.txt"}]}
}
[07/21 17:02:22.359004] INFO: read host from json: 127.0.0.1 .
[07/21 17:02:22.359011] INFO: read user from json: root .
[07/21 17:02:22.359013] INFO: read password from json: ** .
[07/21 17:02:22.408407] INFO: client version: 3.3.6.13Connect mode is : Native[07/21 17:02:22.411254] INFO: Set engine cfgdir successfully, dir:/etc/taos
[07/21 17:02:23.693357] SUCC: host:127.0.0.1 port:0  connect successfully.
[07/21 17:02:23.722231] SUCC: host:127.0.0.1 port:0  connect successfully.
[07/21 17:02:23.722993] SUCC: host:127.0.0.1 port:0  connect successfully.
complete query with 3 threads and 30 sql 1 spend 3.845735s QPS: 7.801 query delay avg: 0.382693s min: 0.367124s max: 0.419617s p90: 0.395527s p95: 0.416581s p99: 0.419617s SQL command: select last_row(*) from meters 
[07/21 17:02:27.569303] SUCC: host:127.0.0.1 port:0  connect successfully.
[07/21 17:02:27.569746] SUCC: host:127.0.0.1 port:0  connect successfully.
[07/21 17:02:27.570143] SUCC: host:127.0.0.1 port:0  connect successfully.
[07/21 17:02:27.572642] INFO: do sleep 1ms ...
[07/21 17:02:27.573148] INFO: do sleep 1ms ...
[07/21 17:02:27.573166] INFO: do sleep 1ms ...
[07/21 17:02:27.574560] INFO: do sleep 1ms ...
[07/21 17:02:27.577159] INFO: do sleep 1ms ...
[07/21 17:02:27.577603] INFO: do sleep 1ms ...
[07/21 17:02:27.577804] INFO: do sleep 1ms ...
[07/21 17:02:27.579129] INFO: do sleep 1ms ...
[07/21 17:02:27.579714] INFO: do sleep 1ms ...
[07/21 17:02:27.581079] INFO: do sleep 1ms ...
[07/21 17:02:27.581663] INFO: do sleep 1ms ...
[07/21 17:02:27.582607] INFO: do sleep 1ms ...
[07/21 17:02:27.583555] INFO: do sleep 1ms ...
[07/21 17:02:27.584650] INFO: do sleep 1ms ...
[07/21 17:02:27.585559] INFO: do sleep 1ms ...
complete query with 3 threads and 30 sql 2 spend 0.016352s QPS: 1834.638 query delay avg: 0.000997s min: 0.000745s max: 0.001392s p90: 0.001370s p95: 0.001387s p99: 0.001392s SQL command: select count(*) from d0 
[07/21 17:02:27.586718] INFO: Spend 5.1750 second completed total queries: 60, the QPS of all threads:     11.594 ,error 0 (rate:0.000%)[07/21 17:02:27.586729] INFO: free resource and exit ...

结果解读：

  • 数据查询测试​：SELECT COUNT(*) FROM d0（​​1,654.6 QPS​​，平均延迟0.000948s ）
  • 数据查询测试： SELECT last_row(*) from METERS（ ​​5.415​​ QPS）， 查询 METERS 表的​​最新行数， METERS 表数据量很大（1 亿行），TDengine 需要扫描索引找到最新数据，可能涉及磁盘 I/O，所以很慢。

四、基准测试结论

  • 写入性能​​：

    • 1亿行数据写入仅耗时25.49秒，吞吐量达 ​​400万行/秒​​ 级别，远超传统时序数据库（如InfluxDB的10万~50万行/秒）。
    • 子表创建速度极快（22,573表/秒），适合大规模设备接入场景。

  • 查询性能​​：

    • COUNT(*)查询​​：QPS极高（1,654.6），因仅需统计单表行数，适合高频监控场景。
    • select last_row()：查询 meters 表的 ​​最新行数据​​（类似于 SELECT * FROM meters ORDER BY ts DESC LIMIT 1），如果 meters 表数据量极大（如 1 亿行），last_row() 需要扫描所有子表或索引，导致延迟高。
http://www.dtcms.com/a/306526.html

相关文章：

  • LeetCode 25：K 个一组翻转链表
  • MCU中的CAN总线是什么？
  • WebRTC核心组件技术解析：架构、作用与协同机制
  • 一文掌握最新版本Monocle3单细胞轨迹（拟时序）分析
  • 如何将JPG、PNG、GIF图像转换成PDF、SVG、EPS矢量图像
  • Rust基础[part9]_返回值和错误处理、模块化
  • [特殊字符] 征服CPU的艺术：Rust多进程编程实战指南
  • Cortex-M处理器的优势？
  • STM32CubeIDE新建项目过程记录备忘（二）
  • FFmpeg：因码流采集与封装不同步导致录制出来的MP4文件会出现黑屏、绿屏的问题
  • Zynq SoC 中断控制系统设计与实现：基于 GPIO 的中断驱动开发
  • LocalDateTime vs Instant vs ZonedDateTime：到底该用哪个？
  • .net6的webapi项目统一封装返回值
  • 剧本杀系统 App 开发：科技赋能，重塑剧本杀游戏体验
  • 光伏气象监测系统：当阳光遇见科技
  • Javascript 基础总结
  • 做题笔记：某大讯飞真题28道
  • 浅拷贝和深拷贝
  • uni-app，uni.navigateTo
  • 【LangChain4j 详解】Java生态大语言模型框架设计哲学与架构原理
  • Node.js以及异步编程
  • vue模块化导入
  • 网络安全学习第16集（cdn知识点）
  • Android调用python库和方法的实现
  • 开源项目：排序算法的多种实现方式
  • DAY15-指针（3）
  • 解决:React Native 中常见的 状态栏遮挡内容
  • python 中 TypeError: self类型对象传入错误解决办法
  • 在职申硕，怎么选适合自己的学科专业呢？
  • 计算机网络1-3：三种交换方式