MySQL 45 lecture learning notes (XIV) count (*)

One .count(*) How to implement

Different MySQL engine ,count（*） There are different implementations

• MyISAM The engine stores the total row number of a table on disk , So execute count（*） It will return this number directly , It's very efficient ;
• InnoDB The engine needs to read it line by line from the engine and count it up

If you add where Conditions ,MyISAM Tables can't return so fast

Two . Why? InnoDB Not put MyISAM Save the data ？

Even in Multiple queries at the same time , because Multi version concurrency control （MVVC）,InnoDB Table should How many lines back I'm not sure .

• InnoDB It's the index organization table , Primary key index tree Leaf nodes are data , And ordinary index tree Leaf nodes are primary key values .
• Ordinary index tree Than Primary key index tree Many small , Yes count(*) This kind of operation , Traverse the index tree The results are the same .

MySQL Optimizer Will find the smallest tree to traverse .

On the premise that the logic is correct , Minimize the amount of data scanned , This is database system design One of the general rules of

3、 ... and .show table status 

show table status command Number of lines displayed also Can't be used directly .

Four . Database summary

• MyISAM surface although count(*) Soon , however Unsupported transaction ;
• show table status command Although it returns quickly , But it's not accurate ;
• InnoDB The table directly count(*) Will traverse the entire table , Although the results are accurate , But it will Cause performance problems .

5、 ... and . Cache system save count

One . Design thinking ：

• Every row inserted in this table Redis Add one to the count
• The count is reduced by one for each Deleted Row
• In this case , Read and update operations are fast

The cache system may lose updates

• A row has just been inserted into the data table ,Redis The value saved in is also added with a line , then Redis Abnormal restart , Restart from the stored redis Read the value back in the place of the data , Just add one and you will lose .

Solution ：

• Redis After abnormal restart , Go to the database and execute it alone count（*） Gets the actual number of rows , Then write the value back to Redis Li will do

In a concurrent system , We are Unable to accurately control the execution time of different threads , Whether it's Write cache first and then database still Write the database first and then the cache There is no guarantee of absolute accuracy .

6、 ... and .count(*)、count( Primary key id)、count( Field ) and count(1) And so on

• about count（ Primary key id） Come on ,InnoDB The engine traverses the whole table , Put each line id out ,server Layer space determination .
• about count(1) Come on ,.server Layer for returning InnoDB The engine traverses the entire table , But every line that does not take value , Put a number “1” go in , Judgment cannot be empty , Add by line .

count(1) Perform better than count( Primary key id) fast . Because back from the engine id It's going to involve parsing the data lines , as well as Copy field values The operation of

about count( Field ) Come on ：

• If this “ Field ” Is defined as not null Words , Read the field line by line from the record , Judgment cannot be null, Add by line ;
• If this “ Field ” The definition is allowed to be null, So when it comes to execution , To judge that it might be null, We need to take out the value and judge again , No null Just add up .

That's the first principle ,server What fields does the layer want ,InnoDB What fields are returned .

count(*) Specially optimized

count（ Field ）<count（ Primary key id）<count（1）==count(*)

recommend count(*)！