当前位置:网站首页>Full link voltage test: the dispute between shadow database and shadow table
Full link voltage test: the dispute between shadow database and shadow table
2022-07-07 05:10:00 【InfoQ】
author :
Ge tianmeng ( Zhi Yun )
What is the popular full link voltage measurement in the industry
Full link voltage measurement was born in Alibaba Shuang 11 Preparation process , If you say double 11 Big promotion is the business of Alibaba “ final exam ”, The full link voltage test is before the big test “ Mock exam ”, After its birth, it is known as double 11 Stability guaranteed “ Nuclear weapons ”. The full link voltage test simulates the high traffic scenario in the production environment , Get the most real online actual carrying capacity 、 Perform accurate capacity planning , Ensure system availability .
The rapid development of distributed architecture and business has brought uncertainty to business systems . Any node in a distributed environment can become a bottleneck / Short board / problem , At the same time, with the rapid growth of business , Facing more severe challenges and uncertainties . such as :
- Single chain road pressure measurement lacks external interference and competition of various resources , The results of single chain road pressure measurement are generally optimistic , It cannot reflect the real system carrying capacity .
- Some problems will only be exposed under real large traffic , Like network bandwidth 、 Impact between systems 、 Basic dependency, etc .
- Full link voltage measurement is more than just pressure measurement , It's more about a real preview , Plan drill 、 Current limiting verification 、 Unified acceptance of high availability schemes such as destructive drills .
The common problem of full link voltage measurement is how to achieve data isolation in the production environment : When carrying out write pressure measurement in the production environment , It is necessary to ensure that the normal operation of online business will not be affected while the pressure test is carried out , Then we need to consider the isolation and storage of the data generated by the pressure test and the real production data , Avoid the impact of dirty data on online business . In addition to providing shadow table solutions, Alibaba cloud's full link voltage testing platform , It also provides the data isolation scheme of shadow database .
In the process of implementing full link voltage measurement in the production environment , For the two schemes mentioned above , Facing the problem of choosing data isolation scheme , This article first aims at shadow database 、 Shadow table two schemes are introduced and compared , Then for common scenarios , Give suggestions on the selection of the scheme .
Selection of full link voltage measurement data isolation scheme
At present, the full link voltage measurement platform provides
Shadow vault 、 Shadow table
etc.
Solution
. How to choose a suitable scheme ? This paper first expounds the principles of the two schemes , Then from the performance 、 stability 、 Compare the three cost indicators .
Scheme 1 : Shadow vault
Pictured 1 Shown , For shadow library solutions , It is to establish the corresponding shadow database on the same instance . After the full link pressure probe mounted by the user service obtains the flow meter, it carries out the corresponding bypass processing , If it is shadow traffic , Then the shadow connection will be obtained from the shadow connection pool for use by the business side , Thus, the data generated by pressure measurement flow is bypassed to the corresponding shadow Library , In order to achieve the effect of data and production library isolation , Thus, the data generated by the pressure measurement flow can avoid polluting the production warehouse .

chart 1: Basic principle of shadow library scheme
Option two : Shadow table
Pictured 2 Shown , Similar to the shadow library scheme , For shadow table scheme , It is to establish the corresponding shadow table on the same database on the same instance . After the full link pressure probe mounted by the user service obtains the flow meter, it carries out the corresponding bypass processing , If it is shadow traffic , that , The probe will be aimed at this DB Call to do SQL Parse and replace , Thus, the data generated by pressure measurement flow is bypassed to the corresponding shadow table .

chart 2: Basic principle of shadow table scheme
Scheme comparison
This article mainly from the performance 、 stability 、 The advantages and disadvantages of the two schemes are described from three aspects of cost .

chart 3: Scheme comparison
performance
Machine specification :4c8g
Concurrency specification : It is necessary to simulate both normal and pressure measured flows , Here we use 2:8 In proportion to , In order to simulate the low peak period of business traffic for the full link voltage test of the production environment .
- Normal flow :200 Concurrent
- Pressure measurement of flow :800 Concurrent
Here we mainly analyze the monitoring of the host where the service is located and the database instance used . among , Application monitoring is mainly based on CPU、 Memory and average RT Analysis of three indicators . The database instance monitors the number of slave connections 、QPS Analyze the dimensions of the two indicators .

It can be seen from the comparison of indicators in different dimensions of the above two schemes , Shadow table scheme is right CPU The consumption is slightly higher , This is related to the implementation of the scheme .
stability
Speaking of stability , You can specify the number of connections from the data source instance 、 Capacity specifications 、IOPS、 Analyze network traffic and other aspects .

The above indicators , Here, take the number of connections as an example , As follows :
For shadow library solutions . Because different databases are established on the same instance , So if you don't consider that the database instance can reach the maximum number of connections , Theoretically, shadow connection and normal connection are independent , Execution does not affect each other .
For shadow table scheme , Because different data tables are established on the same database on the same instance , Here we need to consider the connection pool configuration on the business side , Because shadow traffic involves DB Operation and normal flow are involved DB operation , Database connection used , All come from the same connection pool , So if the pressure measurement level is large , It is easy to have connection bottlenecks in connection pools .
cost

According to the table , Here, the redundancy cost and data migration cost are mainly explained , As follows :
- Redundancy costs
For shadow library solutions , In order to ensure the accuracy of the full link voltage measurement evaluation results , We need to do a full amount of Library migration on the same instance , Including table structure and table data , This will bring about an obvious problem , The cost is relatively high , All basic read-only tables ( This type of table will not have write operations ) There should be one redundant , Unable to achieve the purpose of reuse , So for medium and large enterprises , It's hard to accept .
For shadow table scheme , Is to create a shadow table on the same database on the same instance . Then you can reuse the basic read-only table in the production Library , Just create a shadow table for the write table . The shadow table scheme reduces the cost consumption caused by data redundancy to a certain extent .
- Data migration costs
From the main process of pressure measurement , It is divided into pre pressure measurement 、 Under pressure test 、 After pressure test . among , Data preparation is at the stage before pressure measurement , Whether the pressure test is successful , It is closely related to data preparation . In the process of data migration, other table fields associated with a data table need to be migrated at the same time , This process is more complex and energy-consuming . therefore , Which option to choose , It needs to be evaluated in combination with the complexity of business data .
summary
Sum up , Which of the above two options should be selected , In fact, it is not enough to rely on only one indicator , Comprehensive evaluation should be carried out in combination with the above indicators and specific business scenarios . Here are two typical scenarios to illustrate how to choose a suitable solution , The following comments are for reference only .
scene 1: The proportion of reading tables involved is higher than that of writing tables 、 And the whole database migration cost is high , It is recommended to choose the shadow table scheme , To some extent, it can reduce the cost of complex data migration .
scene 2: The proportion of writing table is higher than that of reading table , At the same time, when the instance capacity of the production library is sufficient , It is recommended to choose the shadow library scheme , It reduces carding to a certain extent 、 Cost of configuration .
at present , Full link voltage measurement provides hybrid cloud solution output , Welcome to use . For details, please click
here
see .
More exchanges , Welcome to the nail group to communicate ,PTS User AC nail group number :11774967.
边栏推荐
猜你喜欢
随机推荐
创始人负债10亿,开课吧即将“下课”?
Batch normalization (Standardization) processing
DBSync新增对MongoDB、ES的支持
AttributeError: module ‘torch._ C‘ has no attribute ‘_ cuda_ setDevice‘
精彩速递|腾讯云数据库6月刊
[736. LISP syntax parsing]
Stm32f103ze+sht30 detection of ambient temperature and humidity (IIC simulation sequence)
带你遨游银河系的 10 种分布式数据库
高数中值定理总结
c语言神经网络基本代码大全及其含义
Dynamically generate tables
为什么很多人对技术债务产生误解
Dbsync adds support for mongodb and ES
STM32F103实现IAP在线升级应用程序
File upload vulnerability summary
AOSP ~Binder 通信原理 (一) - 概要
JS 的 try catch finally 中 return 的执行顺序
STM32 system timer flashing LED
Thread和Runnable创建线程的方式对比
ThinkPHP关联预载入with



![[736. LISP syntax parsing]](/img/62/5e2aeec150096aa3fd81025d146255.png)




