Why does the SQL statement hit the index faster than it does not?

If you need a full version of the interview document, you can scan the QR code at the bottom of the article to get it for free ！

A fan was asked when he was interviewing Gao Kai , Why? SQL Statements hit the index faster than they miss the index ？ Although I know the answer , But the moment I was asked , I don't know how to organize language . today , Let me give you an in-depth analysis .

1、 The function of index

Imagine , Now there is a dictionary containing hundreds of thousands of words , Hundreds of pages thick , At the same time, the words inside are arranged in disorder . If you do not use a directory , How can we find the words we need from the dictionary ？ without doubt , We can only turn page by page , obviously , This is an anti human work .

What we must think is to look at the catalogue first , then , Find the relevant word or radical , then , Find the corresponding page number and then find the text you want to find , such , The efficiency is greatly improved . As a matter of fact , A directory is an index , The idea of database index that we say comes down in one continuous line with the idea of directory .

The main function of database index is to help us quickly retrieve the data we want , Thus, it is not necessary to do a global scan for each query .

Assuming that no algorithm is used , We need to check 10 One of ten thousand records , In the worst case, you need to traverse 10 Ten thousand times .

But if you use the binary search algorithm , We just need to do log2 20000 Time , That is to say 14.287712 Next time . This means that we only need to sort the sorted values 14 Search , You can use bisection to find the unique value you want , Common index data structures are B Trees and B+ Trees .

Next we , With MySQL Of InnoDB Engine as an example , Analyze how indexes work .

2、 Index execution principle

We know MySQL Of InnoDB The engine uses B+ Tree data structure , When we go to execute SELECT Statement to query data ,InnoDB Need to read data from disk , And this process will involve disk And random disk IO , Let's take a look at this picture ：

The system will transfer the logical address of the data to the disk , Disk control circuit translates logical address into physical address according to addressing logic . That is to determine which track the data to be read is on 、 Which sector? . In order to read the data of this sector , You need to put the head on this sector , In order to achieve such a point , The disk will continue to rotate . Rotate the target sector under the head , So that the head can find the corresponding track . There is also a loss of seek time and rotation time . Obviously disk IO The performance overhead of this process is very large , Especially when there is a large amount of data to query .

therefore InnotDB Inside , Simply build an index on the data stored on the disk , Then, the index data and the disk address corresponding to the index column are marked with B+ Tree storage . Look at this picture ：

When we need to find the target data , According to the index from B+ Just find the target data in the tree . because B+ There are many subtrees in the tree , therefore , Only a few disks are required IO You can find the target data .

as for B+ The data structure of the tree , There is no analysis here . You can go to my personal home page to see the videos in previous periods .

3、 Disadvantages of index

although , Using indexes can reduce the number of disks IO frequency , Improve query efficiency , however , Indexes can't be built too much . If the indexes of all fields in a table are large , It also leads to performance l falling . Imagine , If an index is as long as a table , Then it will again become an overhead to check . It's like a dictionary with a very detailed catalogue , But its length is as long as all words , At this time, the efficiency of the directory itself is greatly reduced .

Is there any defect in the index ？ There must be , Indexing can improve query read performance , And it will reduce write performance . When there is an index , If you change a record , Or insert a new record into the database , It will perform two write operations （ One operation is to write the record itself , Another operation is to update the index ）.

therefore , When defining an index , The following must be kept in mind ：

1、 Each field in the index table will reduce write performance .

2、 It is recommended to index fields with unique values in the table .

3、 Fields that act as foreign keys in a relational database must be indexed , Because they facilitate complex queries across multiple tables .

4、 Indexes also use disk space , Therefore, be careful when selecting fields to index .

   Pay attention to WeChat public number 『 Tom Bomb architecture 』 reply “666” Available 200 page pdf Format document （ Complimentary gift 10 Internal interview materials of Wanzi factory ）. Thank you for your support ！

You can also scan the QR code below +\/ Get the full version free of charge ！

↓ ↓ ↓ ↓ ↓