Linux server development, MySQL index principle and optimization

Recommend a free open course of zero sound College , Personally, I think the teacher spoke well , Share with you ：Linux,Nginx,ZeroMQ,MySQL,Redis,fastdfs,MongoDB,ZK, Streaming media ,CDN,P2P,K8S,Docker,TCP/IP, coroutines ,DPDK Etc , Learn now

1. summary

• Indexes are physical concepts , Constraints are logical concepts .
• The difference between a primary key index and a unique index is that the primary key index cannot be empty , A unique index can have one NULL value .
• Component table PRIMARY KEY Create a primary key for the display , If it is not created, it will be in UNIQUE NOT NULL Select the first one as the primary key . If there is no non empty unique index ,MySql Then automatically generate 6 byte _rowid A primary key .
• A foreign key constraint is a constraint , To ensure the integrity of the two tables , It's transactional , Try not to use .
  

2.B+ Trees

The full name of binary tree is balanced binary search tree ,B+ Tree is a multi-path balanced search tree , A characteristic of the search tree is that traversal in middle order is an ordered structure . The balance is the height of the tree , The aim is to provide a stable search time complexity , It is the time complexity that remains stable after adding, deleting, modifying and checking . If it becomes a linked list, it will become O(n) Time complexity of , Is not a balanced .B+ The tree is a very balanced , It is mainly reflected in that leaf nodes are on the same layer , The height of each link is the same .
The height of the tree represents the number of comparisons , Red and black trees are tall and thin , Used in memory ; and B+ The tree is short and plump , Used to access disks . One disk IO yes 10ms about , A memory access 100us about , Millions of differences .
B+ How the tree maps to the disk ？ A node is 16K The data of , When accessing disk data, the size of the data pulled is usually 4K or 8K, It's exactly an integral multiple , The configuration file can also be modified .B+ One node of the tree must store at least two rows of data . If you exceed 16K, The excess will be mapped to the shared table space . Non leaf nodes store index information , Leaf nodes store data information .

B+ Tree range query LIMIT1,5, Adjacent leaf nodes refer to each other , Avoid backtracking queries .
Auto increment can be used if the data is large bigint type , If every second 1 Billion data , need 5849 It takes years to run out ！
myisam in frm At the end is the table information file ;myd It's a data file , Use heap table to organize storage ;myi Is the index file , use B+ Tree organization storage .
inodb in frm Is the table information structure ;bbd It's a data file , use B+ A clustered index created by a tree primary key . When using a secondary index to find , Will find the primary key id, Through the primary key id Find row data , This is called a return table query .
inodb By segment 、 District 、 Page composition , Segments are divided into data segments , Index segment , Rollback segments, etc ; The size of the area is 1MB（1 Zones by 64 Composed of consecutive pages ）; The default value of the page is 16K; Page is a logical page , The physical size of the disk is generally 4K perhaps 8K; In order to ensure the continuity of pages in the area , The storage engine usually requests from the disk at one time 4-5 Districts .

• Buffer pool
• Change buffer: Cache non unique indexes DML Add, delete and change operation , The cache first . If it is a unique index, write the log first , Then the asynchronous thread brushes the disk , The non unique index will be written to the cache and then flushed .
• Log buffer
• LRU The algorithm switches out the rarely used data , This is the cached hot data policy . The ring buffer has no data movement .

Left most matching principle

For composite indexes , Match from left to right , encounter <>between like Just stop matching .

Overlay index

Data can be found through the secondary index , Instead of using a clustered index . The secondary index data contains less , Therefore, the tree height is generally lower than that of the clustered index , So there will be fewer disks io. Query clustered indexes and auxiliary indexes will not be queried back to the table .

Index failure

• select … where A and B if A and B One of the does not contain an index , The index becomes invalid
• The index field participates in the operation , The index becomes invalid .from_unixtime(idx)=‘2021-04-30’;
• Implicit conversion of index field , The index becomes invalid .‘1’ Implicitly convert to 1;
• LIKE Fuzzy query , wildcard % start , The index becomes invalid .select *from user where name like ‘%ark’;
• Use... On index fields NOT <> != Index failure .id<>0, Then change to idx>0 or idx<0;
• In the composite index , The first column index is not used , The index becomes invalid .