当前位置:网站首页>Tiflash performance tuning
Tiflash performance tuning
2022-07-19 11:41:00 【Tianxiang shop】
This article introduces how to make TiFlash Several ways to achieve optimal performance , Including planning machine resources 、TiDB Parameter tuning 、 To configure TiKV Region Size, etc .
Resource Planning
For those who want to save machine resources , And there is no isolation requirement at all , have access to TiKV and TiFlash Joint deployment . The suggestion is TiKV And TiFlash Leave enough resources separately , And don't share disks .
TiDB Related parameter tuning
about OLAP/TiFlash exclusive TiDB node , It is recommended to increase the number of read concurrency tidb_distsql_scan_concurrency To 80:
set @@tidb_distsql_scan_concurrency = 80;Turn on Super batch function :
tidb_allow_batch_cop Variable is used to set from TiFlash When reading , Whether to put Region Request to merge . When the query involves Region The quantity is relatively large , You can try to set this variable to
1( ContrabandaggregationPush down to TiFlash Coprocessor Your request is valid ), Or set the variable to2( Push all down to TiFlash Coprocessor The request is effective ).set @@tidb_allow_batch_cop = 1;Try to turn on aggregation push
Join/Unionetc. TiDB Operator optimization :tidb_opt_agg_push_down Variable is used to set whether the optimizer performs aggregate function push down to Join Previous optimization operations . When aggregation operations in queries are slow , You can try to set this variable to 1.
set @@tidb_opt_agg_push_down = 1;Try to turn on
DistinctPush throughJoin/Unionetc. TiDB Operator optimization :tidb_opt_distinct_agg_push_down Variable is used to set whether the optimizer executes with
DistinctAggregate function of ( such asselect count(distinct a) from t) Push down to Coprocessor Optimized operation of . When the query containsDistinctWhen the aggregation operation of is slow , You can try to set this variable to1.set @@tidb_opt_distinct_agg_push_down = 1;
边栏推荐
- C # build a system based on WPF entry project of net5
- 下推计算结果缓存
- Antd form setting array fields
- Ten minutes from pytorch to mxnet
- Keras深度学习实战(14)——从零开始实现R-CNN目标检测
- 8. Fixed income investment
- Research on Wenhua commodity index
- Learning note 3 -- basic idea of machine learning in planning control
- 翻墙后看什么?最热门的国外网站——翻墙网址导航
- 梦想CMS 前台搜索SQL注入
猜你喜欢

Introduction and Simulation Implementation of string class

性能优化之@Contended减少伪共享

Leetcode 1328. 破坏回文串(可以,已解决)

TCP congestion control details | 7 Surpass TCP

02-3、指针和引用的区别

Total number of blocking and waiting in jconsole thread panel (RPM)

Sword finger offer II 041 Average value of sliding window

【多线程】JUC详解 (Callable接口、RenntrantLock、Semaphore、CountDownLatch) 、线程安全集合类面试题

【嵌入式单元测试】C语言单元测试框架搭建

【二叉树】之力扣牛客必刷题
随机推荐
[unity technology accumulation] realize the mouse line drawing function &linerenderer
A current List of AWESOME Qt and qml
Cmake common commands (V)
function/symbol ‘pango_context_set_round_glyph_positions‘ not found in library ‘libpango-1.0.so.0‘x
Learning outline of the column "MySQL DBA's magic road"
Will causal learning open the next generation of AI? Chapter 9 Yunji datacanvas officially released the open source project of ylarn causal learning
At5147-[agc036d]negative cycle [DP, model conversion]
Conversion of unity3d model center point (source code)
03-2、
JVM hook hooks function
A simple websocket example
02-2、缺省参数、函数重载、引用、隐式类型转换、关于报错
Chapter 1 of creating virtual machine (vmvare virtual machine)
Learning note 3 -- basic idea of machine learning in planning control
Mpu9250 ky9250 attitude, angle module and mpu9250 MPL DMA comparison
Kernel mode and user mode
The difference between CPU load and CPU utilization
jconsole线程面板中的阻塞总数和等待总数(转)
动态内存分配问题
下推计算结果缓存