Several optimization scenarios using like fuzzy retrieval in SQL

SQL We often encounter the use of LIKE The scene of fuzzy retrieval ,'%' The location of , It may affect the normal use of the index , I saw an article in Teacher Liu's official account , The transformation strategies of relevant scenes are introduced , Very useful , Recommended reading .

The test table t1,object_name Column creation index ,

create table t1 as select * from dba_objects;
create index idx_t1_01 on t1(object_name);

Scene one ,'%' After

select object_name from t1 where object_name like 'BISAL%';

Obviously, you can use indexes , But be careful , In this case ,'%' The shorter the first string , The less selective the index is ,

Scene two ,'%' before

select object_name from t1 where object_name like '%BISAL'

Because this index is based on object_name The positive order of columns is organized in an index , Fuzzy search of header , Cannot locate data directly through the index , Just because the search column only has object_name, So the index fast full scan is used , Actually, it is all the leaf nodes scanned ,

To make it clear , We search for object_id Column , He is not in the index ,

select object_id from t1 where object_name like '%BISAL';

therefore ,Oracle Lower cost full table scan selected ,

As a comparison , We are in accordance with the 'BISAL%', The scenario test that can use the index ,

select object_id from t1 where object_name like 'BISAL%';

You can see , Index range scanning is used , obtain rowid, Go back to the table to get specific data , There is no need to scan the entire index or the entire table ,

Let's pull it back , If you have to '%BISAL' retrieval , You can create a object_name Reverse index of columns ,

create index idx_t1_02 on t1(reverse(object_name));

In the query statement LIKE The right value of is also used reverse function ,

select object_name from t1 where reverse(object_name) like reverse('%BISAL');

here ,'%BISAL' Index is used , Careful friends may find that the implementation plan is slightly different from the above , There are many operations for returning tables , The reason is that the index is based on reverse(object_name) The organization's , But the search is object_name, Therefore, the table should be returned according to the index ,

Scene three , Before and after '%'

for example '%BISAL%', Can I use an index ？

There are three situations ,

(1) ABC Always appears at a fixed position starting with a string , You can create functional indexes to optimize .

(2) ABC Always appear at the end of a string , You can create a function composite index to optimize .

(3) ABC The position in the string is not fixed , It can be done by rewriting SQL To optimize .

Case one ,ABC Always appears at a fixed position starting with a string .

Can pass substr Function to intercept a string , Create function indexes .

for example ,BISAL From the fifth bit of the string , Create function indexes ,

create index idx_t1_03 on t1(substr(object_name, 5, 30));

The implementation is as follows SQL, It is equivalent to intercepting from the fifth bit of the original string every time ,

select object_name from t1 where substr(object_name, 5, 30) like 'BISAL%';

You can use indexes ,

The second case ,ABC Always appear at the end of a string , You can create a function composite index to optimize .

It is equivalent to the need to intercept strings in reverse order , Can pass reverse and substr Composite function index , for example BISAL From the last five digits of the string , Create function indexes ,

create index idx_t1_04 on t1(reverse(substr(object_name, 1, length(object_name)-4)));

When searching , Need to use reverse and substr Function combination ,like Right value '%BISAL', Can be realized ‘%BISAL%’ Search function ,

select object_name from t1 where reverse(substr(object_name, 1, length(object_name)-4)) like reverse ('%BISAL');

The third case ,ABC The position in the string is not fixed , It can be done by rewriting SQL To optimize .

This needs to be rewritten , hypothesis object_name There is an index , The requirements are as follows ,

select object_name from t1 where object_name like '%BISAL%';

We rewrite it as , Through a subquery , And conditions object_name relation ,

select object_name from t1 where object_name in (select object_name from t1 where object_name like '%BISAL%');

The execution plan at this time , As shown below , A combination of index fast full scan and index range scan ,

Even if we search object_id This field that is not in the index ,

select object_id from t1 where object_name in (select object_name from t1 where object_name like '%BISAL%');

It also avoids full table scanning , Although we still need a quick full scan of the index , But at least the cost of scanning is reduced (1/N( The ratio of the number of index blocks to the number of data blocks )), On this alone , The bigger the watch , The more obvious the effect is . But this kind of IN rewrite , If the subquery returns a small number of records , Execution efficiency may be improved N times , But if more , Rewriting efficiency , It may not be much different from before ,

Although the above '%' There are various transformation schemes , But at least some of them have to be rewritten SQL, So I suggest , From the demand level , Be sure to use LIKE Is the scene of fuzzy retrieval reasonable , Whether his non functional indicators meet the requirements , Don't come up and change , Plan before you act , You may get twice the result with half the effort .

Recently updated articles ：

《 Short track speed skating and speed skating 》

《InnoDB perform delete What did you do ？》

《 Recent problems 》

《 Reading notes on Data Governance - Of enterprise data governance “ Avenue 、 Law 、 Technique 、 device ”》

《 clone PDB Database operation 》

Article classification and indexing ：

《 official account 900 Article classification and index 》