Master-slave separation of database

It seems that the master database has been separated from the Internet service , Except for start-up companies, they may not adopt... When the business volume is small , The rest will be a master-slave separation by default . Today, let's talk about the details of the separation of master and slave .

First of all, master-slave separation generally refers to read-write separation , The main library undertakes the task of writing , The Reading Library undertakes the task of reading . A service open to users often reads far more than writes , But the pressure of database reading is often greater than that of writing , So here we will write the library （ It is generally called the main library ） And read Library （ It is generally called slave library ） Separate , It can greatly improve the performance of the business system . But there is no real separation between reading and writing , The data still needs to be copied from the master database to the slave database , Is only for api The request for , The read request will fall on the slave Library , The written request falls on the main library .

With mysql For example , Explain how the master-slave replication of data is realized .

A simple diagram is drawn to show the process of master-slave synchronization . First, data changes will be recorded log,mysql This is called binlog, When the slave library is set , An asynchronous... Will be created from the library IO Thread to read the main library binlog, After arriving at the library, it became relaylog, From here log It can be interpreted as SQL, Finally, enter the slave Library . Because it's an asynchronous thread , And binlog The changes were subscribed from the library , Database can be considered as quasi real-time synchronization , But the delay must be better than that of a single machine mysql Will send some .

From stand-alone to cluster , The complexity of the system has increased , There will certainly be some additional problems . This is mainly about 2 A question .

As mentioned in the previous chapter , There must be delay in the synchronization between master and slave , After all, it's a network connection , Speed depends on network speed 、 Packet size 、 Request frequency, etc . For some scenarios , Even if the master-slave synchronization is slow , For example, I placed an order in the e-commerce system, but I didn't find it in my order for the time being , Then after a few seconds, I found , This is not an unacceptable problem , It's not even a problem in the eyes of many users . But in another case , It may be a big problem . Let's take e-commerce as an example , This time the protagonist becomes a programmer . In the code save An order data , Then go to another place to query the data of the order to process further business , At this time, because the delay is not found, it will feel that this order does not exist , And then produce more serious consequences .

What is the solution to this problem ？

At present, the mainstream practice in the market is to give mysql Be another agent , This layer is not visible to business developers , Will automatically identify SQL, Then do some rewriting to access different libraries . The existing ones are 360 Of Atlas And meituan are based on Atlas Further developed DBproxy. This will increase 1 Secondary network situation , It will have a slight impact on performance .

Another is to wrap at the code level , Let the business code decide which library to connect to . This approach is not universal , Every language should be implemented once , When you have a whim to try a new programming language , This is the real stumbling block . But just stick to it honestly Java Position , This is also a higher performance approach .

The separation of master and slave is not mysql Peculiar , This is actually a solution to the problem . Like a tidal driveway , Different scheduling strategies are given according to the scenario . stay redis and es Such storage is also widely used .