Distributed transaction -- middleware of TCC -- selection / comparison

Original website ： Distributed transactions --TCC Middleware -- The selection / contrast _IT A blog with a sharp blade -CSDN Blog

brief introduction

explain

        This article introduces the TCC Some of the middleware . Include ：Seata,TCC-transaction、ByteTCC、Himly、EasyTransaction、LCN

Why use TCC middleware

        Perceive the implementation of each stage and push forward the next stage of these things , It's impossible to do it by hand , It's too complicated .

Middleware introduction

Middleware name	Github Address	star Number	brief introduction
Seata	https://github.com/seata/seata	18.3K	Ali .TCC Mode only supports dubbo, I won't support it feign.
tcc-transaction	https://github.com/changmingxie/tcc-transaction	4.9K	personal . I won't support it SpringBoot！！
LCN	https://github.com/codingapi/tx-lcn	3.9K	CodingApi The team . Too much dependence ：Redis、Zookeeper、Consul; at present (2021.02.24) Not yet TCC Pattern .
Hmily	https://github.com/dromara/hmily	3.5K	Country garden . Well documented ,SpringCloud And SpringBoot support , There is a lot of information on the Internet .
ByteTCC	https://github.com/liuyangming/ByteTCC	2.5K	Beijing Xinao Group
EasyTransaction	https://github.com/QNJR-GROUP/EasyTransaction	2.2K	personal .（ Lack of information 、 Fewer documents ）

forecast ： Hmily Of star The number will gradually equal and surpass LCN and tcc-transaction.

Functional comparison

Frame name	Idempotency	Nested calls	RPC Framework support	SpringBoot Support	Supported transaction logs
Seata			Dubbo.（springcloud：AT Mode support ,TCC Mode not supported ）	Support	file、DB、redis
tcc-transaction	I won't support it	I won't support it	Uncoupled RPC frame . namely ： The bottom layer uses any RPC Fine . But actually ： Yes. dubbo Support .	I won't support it	file、DB、redis、ZK
LCN	I won't support it	I won't support it		Support .（ Need to be 5.0 Above version ）
Hmily	I won't support it	I won't support it	Dubbo、motan、springcloud		file、DB、redis、mongodb、ZK
ByteTCC	I won't support it	I won't support it	Dubbo、springcloud		file
EasyTransaction	Support	Support	Dubbo、SpringCloud、Ribbon/Eureka		DB、Redis

explain ： It doesn't matter whether idempotent and nested calls are supported . reason ：

• In our business, we usually deal with idempotence by ourselves （ Whether or not the framework supports idempotence ）;
• If the framework does not support nested calls , We don't nest calls .

LCN

explain

LCN TXC TCC Three transaction modes , And mixed support .

LCN It's essentially a BestEffors 1PC Framework ： Most of the time , As long as the application does not Crash It will not lead to inconsistency .

advantage

• Excellent performance
• High reliability
• The code is less intrusive
  In all the frameworks of this comparison ,LCN Implementation of distributed transaction processing mode , The coding complexity and the amount of intrusion code are the lowest .

shortcoming

• Additional deployment required tx-manager Service node .
• Because of the need to lock Resources are handled in this way , If you update some popular products ,LCN The performance degradation of is greater than TCC Pattern .
• Service timeout , It will cause the resources of other services to be locked , For example, in the process of payment service timeout , The inventory of relevant goods will always be inoperable .
• It can easily lead to inconsistent data ： For data inconsistency , Conditions that must be repaired
  • We have to write the corresponding repair program . This is actually similar to TCC/ Compensation and other workload are the same
  • Use BestEffors1PC When data exceptions occur in the business code written , There is no guarantee that its subsequent compensation is enforceable .
  • for instance , Transfer accounts , There's money for others , But the data that he didn't deduct money was abnormal , At this point, the two customers may immediately withdraw their money and use it

EasyTransaction

advantage

• Multi thread concurrent processing mode is adopted for remote calls , Better performance optimization
• High reliability
• Simple integration , There is no need to deploy additional transaction management nodes

shortcoming

• Compared to other frameworks , The intrusion and coding amount are too large
• RPC Request timeout processing has not been implemented

hmily

explain

“ use disruptor The framework reads and writes transaction logs asynchronously .

advantage

Declare in the form of annotations TCC Of course confirm And cancel Method , The code is less intrusive .

shortcoming

• The locking mechanism of multithreading needs to be optimized .
• In distributed transactions RPC Request serial processing , The response time of the request is long
• unreliable ： Writing the transaction log asynchronously is equal to TCC It's not reliable

easy-transaction

advantage

• Really high performance , Yes IO A lot of optimization , When multiple independent third-party companies evaluate the distributed transaction framework horizontally , The performance is the best under the same scenario and the same reliability guarantee （ Self validating test ）
• Theoretically, as long as the external components do not lose data , stay ET There will be no incomplete transactions inside （ Relative to the above frameworks , Its principle is not reliable , Abnormal operation is inevitable ）
• Supports frame idempotent , Business processes no longer need to take over idempotent 、 Call tedious and repetitive logic such as timing disorder processing （ This is a tedious and repetitive work , It seems that there is only ET This item is fully supported ）
• Implementation based on extension , Instead of a modified implementation , Easier to understand , Less risk , More powerful
• Support multiple transaction forms （TCC, compensate , Reliable information ,SAGAs etc. ） A mixture of , The most appropriate transaction form can be selected according to the most suitable business （ At this time ET The concept and characteristics at the beginning of its establishment , It is also a feature that other frameworks do not have ）

shortcoming

• There are few documents