Mysql5 available clusters

1. summary

 

We are thinking about MySQL High availability architecture of database , We should mainly consider the following aspects ：

• If the database is down or accidentally interrupted , Can restore database availability as soon as possible , Reduce downtime as much as possible , Ensure that the business will not be interrupted due to database failure .
• For backup 、 The data of non primary node with read-only copy and other functions should be consistent with the data of primary node in real time or finally .
• When a database switch occurs in the business , The database content before and after the switch should be consistent , It will not affect the business because of missing or inconsistent data .

We won't discuss the high availability rating in detail here , This paper only discusses the advantages and disadvantages of common high availability solutions and the selection of high availability solutions .

 

 

2. High availability solution

 

2.1. Master slave or master master master semi synchronous replication

 

Using a two node database , Set up one-way or two-way semi synchronous replication . stay 5.7 In later versions , because lossless replication、logical The introduction of some new features such as multithreading replication , bring MySQL Native semi synchronous replication is more reliable .

 

Common architectures are as follows ：

Usually with proxy、keepalived Third party software is used at the same time , It can be used to monitor the health of the database , It can also execute a series of administrative commands . If the main warehouse fails , After switching to the standby database, the database can still be used .

 

advantage ：

1. The architecture is relatively simple , Using native semi synchronous replication as the basis for data synchronization ;
2. Two knots , There is no host selection problem after the host is down , Switch directly ;
3. Two knots , Less resources are needed , Simple deployment ;

 

shortcoming ：

1. Completely dependent on semi synchronous replication , If semi synchronous replication degenerates to asynchronous replication , Data consistency cannot be guaranteed ;
2. Need extra consideration haproxy、keepalived High Availability Mechanism of .

 

 

2.2. Semi synchronous replication optimization

 

The semi synchronous replication mechanism is reliable . If semi synchronous replication is always in effect , Then the data can be considered consistent . But because of some objective reasons such as network fluctuation , Causes the semi synchronous replication to time out and switch to asynchronous replication , Then the consistency of the data cannot be guaranteed . So try to ensure semi synchronous replication as much as possible , Then you can improve the consistency of the data .

 

The scheme also uses a two node architecture , But on the basis of the original semi identical replication, the function is optimized , Make the mechanism of semi synchronous replication more reliable .

 

The reference optimization scheme is as follows ：

 

2.2.1. Dual channel replication

Redeposit failed Upload again Cancel

Semi synchronous replication due to timeout , Copy break , When replication is established again , Set up two channels at the same time , One of the semi synchronous replication channels starts at the current location , Ensure that the slave machine knows the progress of the current host execution . The asynchronous machine starts by catching up on another data channel . When the asynchronous replication channel catches up to the start of semi synchronous replication , Resume semi synchronous replication .

 

2.2.2. binlog File server

Build two semi synchronous replication channels , The semi synchronous channel connecting to the file server is not enabled normally , When the master-slave semi synchronous replication network problems degenerate , Start the semi synchronous replication channel with the file server . When the master-slave semi synchronous replication is restored , Close the semi synchronous replication channel with the file server .

 

advantage ：

1. Two knots , Less resources are needed , Simple deployment ;
2. Simple architecture , There's no election question , Switch directly ;
3. Compared to native replication , The optimized semi synchronous replication can ensure the consistency of data .

 

shortcoming ：

1. Need to modify the kernel source code or use mysql Communication protocol . Need to have a certain understanding of the source code , And can do a certain degree of secondary development .
2. Still rely on semi synchronous replication , There is no fundamental solution to the problem of data consistency .

 

 

2.3. High availability architecture optimization

 

Extend dual node database to multi node database , Or a multi node database cluster . You can choose one master and two slaves according to your own needs 、 A cluster with one master and many slaves or a cluster with multiple masters and multiple slaves .

 

Because of semi synchronous replication , There is a feature that the successful semi synchronous replication is considered successful after receiving a successful response from a slave , Therefore, the reliability of multiple semi synchronous replication is better than that of single semi synchronous replication . And the probability of multi node downtime is less than that of single node , Therefore, to some extent, multi node architecture can be considered as better than dual node architecture .

 

But because of the large number of databases , So we need database management software to ensure the maintainability of the database . You can choose MMM、MHA Or various versions of proxy wait . Common solutions are as follows ：

 

2.3.1. MHA+ Multi node cluster

MHA Manager It will detect the master node , When master Failure time , It can automatically send the latest data to slave Upgrade to a new master, And then put all the other slave Point back to the new master, The entire failover process is completely transparent to the application .

 

MHA Node Run on each MySQL Server , The main function is to process binary log when switching , Make sure to switch as little data as possible .

 

MHA It can also be extended to the following multi node clusters ：

advantage ：

1. It can detect and transfer faults automatically ;
2. Good scalability , It can be expanded as needed MySQL The number and structure of the nodes ;
3. Compared to the two node MySQL Copy , Three nodes / Multinode MySQL The probability of unavailability is lower

 

shortcoming ：

1. At least three nodes are needed , It needs more resources than two nodes ;
2. The logic is more complicated , Troubleshooting after failure , Positioning is more difficult ;
3. Data consistency is still guaranteed by native semi synchronous replication , There is still a risk of data inconsistency ;
4. It may be because of the split brain phenomenon in the network partition ;

 

 

2.3.2. zookeeper+proxy

 

Zookeeper Using distributed algorithm to ensure the consistency of cluster data , Use zookeeper Can effectively guarantee proxy High availability , Can better avoid network partition phenomenon .

advantage ：

 

1. Better guarantee the high availability of the whole system , Include proxy、MySQL;
2. Good expansibility , It can be expanded into large-scale clusters ;

 

shortcoming ：

1. Data consistency still depends on native mysql Semi-synchronous replication ;
2. introduce zk, The logic of the whole system becomes more complicated ;

 

 

2.4. Shared memory

 

Shared storage realizes the decoupling of database server and storage device , Data synchronization between different databases no longer depends on MySQL The native copy function of , But by means of disk data synchronization , To ensure data consistency .

 

2.4.1. SAN Shared storage

 

SAN The concept of processors and storage devices is allowed to （ The server ） Between the establishment of a direct high-speed network （ And LAN comparison ） Connect , Through this connection, data can be stored in a centralized way . The common architectures are as follows ：

 

When using shared storage ,MySQL The server can mount the file system and operate normally , If the master database goes down , The standby database can mount the same file system , Ensure that the master database and the standby database use the same data .

advantage ：

1. Two nodes , Simple deployment , The switching logic is simple ;
2. Good to ensure strong consistency of data ;
3. Not because MySQL The data is inconsistent due to the logic error of ;

 

shortcoming ：

1. High availability of shared storage needs to be considered ;
2. Expensive ;

 

 

2.4.2. DRBD Disk copy

 

DRBD It's software based 、 Network based block replication storage solution , It is mainly used for disks between servers 、 Partition 、 Logical volume and so on , When a user writes data to a local disk , It also sends data to the disk of another host in the network , Such a local host ( Master node ) With the remote host ( For the node ) To ensure real-time synchronization . The common architectures are as follows ：

When something goes wrong with the local host , A copy of the same data remains on the remote host , Can continue to use , Data security is guaranteed .

 

DRBD yes linux Fast level synchronous replication technology implemented by kernel module , It can be done with SAN Achieve the same shared storage effect .

 

advantage ：

1. Two nodes , Simple deployment , The switching logic is simple ;
2. Compared with SAN Storage network , The price is low ;
3. Ensure strong consistency of data ;

 

shortcoming ：

1. Yes io High performance impact ;
2. No read operation is provided from the library ;

 

 

 

2.5. Distributed protocol

 

Distributed protocol can solve the problem of data consistency . The common solutions are as follows ：

 

2.5.1. MySQL cluster

 

MySQL cluster It is the deployment scheme of the official cluster , By using NDB The storage engine backs up redundant data in real time , Realize the high availability and data consistency of database .

advantage ：

1. All using official components , Don't rely on third-party software ;
2. Can achieve strong consistency of data ;

 

shortcoming ：

1. Less used in China ;
2. Complex configuration , Need to use NDB Storage engine , And MySQL There are some differences in conventional engines ;
3. At least three nodes ;

 

 

2.5.2. Galera

 

be based on Galera Of MySQL High availability cluster , It's multi master data synchronization MySQL Clustering solutions , Easy to use , No single point of failure , High availability . Common architectures are as follows ：

 

advantage ：

1. Multi master write , No delay replication , It can ensure strong data consistency ;
2. There is a mature community , There are Internet companies using it on a large scale ;
3. Automatic failover , Automatic addition 、 Remove nodes ;

 

shortcoming ：

1. Need to be native MySQL Node beating wsrep Patch
2. Only support innodb Storage engine
3. At least three nodes ;

 

 

2.5.3. POAXS

 

Paxos The algorithm solves the problem of how a distributed system works with respect to a certain value （ The resolution ） To reach an agreement . This algorithm is considered to be the most effective of its kind .Paxos And MySQL It can be realized in distributed MySQL Strong consistency of data . Common architectures are as follows ：

advantage ：

1. Multi master write , No delay replication , It can ensure strong data consistency ;
2. There is a mature theoretical basis ;
3. Automatic failover , Automatic addition 、 Remove nodes ;

 

shortcoming ：

1. Only support innodb Storage engine
2. At least three nodes ;

 

 

3. summary

 

With the increasing demand for data consistency , More and more methods are being used to solve the problem of distributed data consistency , Such as MySQL Self optimization 、MySQL Optimization of cluster architecture 、Paxos、Raft、2PC The introduction of algorithm and so on .

 

And using distributed algorithms to solve MySQL Database data consistency problem method , More and more people accept it , A series of mature products such as PhxSQL、MariaDB Galera Cluster、Percona XtraDB Cluster More and more are being used on a large scale .

 

With the official MySQL Group Replication Of GA, Using distributed protocol to solve the problem of data consistency has become the mainstream direction . Expect more and more excellent solutions to be presented ,MySQL High availability problems can be better solved .

 

reference

 

[2015 OTN] Peng Lixun -DoubleBinlog programme .pdf