Oracle hint understanding

hint know

The correct grammar is ：

select /*+ index(x idx_t) */ * from t x where x.object_id=123

/*+ */ It's like a comment , There is one more comment “+”, This is it. Hint

Above this hint It means to make Oracle Execute this SQL Forced index .

If hint There is a syntax error in ,Oracle No mistake , Just put /* */ The content in is just a comment .

Unreasonable use Hint Hazards of ：

Because the data in the table will change , Generally, it cannot be in the program sql In the use Hint, If it's like the one above Hint The same forced index . In case one day object_id=123 The return result of takes up of the whole table 50% above , At this time, the index will be slower than the full table scanning . Therefore, we should not force all cases to follow the index .Hint Generally used for one-time execution , For example, do data extraction . And it's average Oracle stay 99% Will judge whether it is correct to go to the index , We don't need to specify .Hint Just to cope 1% Under the circumstances .

Append Use

append It's another kind of Hint, General usage ：

insert /*+ append */ into b select * from a;

such insert Than ordinary insert It's going to be faster , But it's also costly .

1、 When the data in the table is delete in the future , Table spaces leave gaps , The next time insert Will go to fill the gap . however append Of insert Won't look for space , And directly added to the new space . If you use it all the time append, It will make the table space larger and larger .

2、 This point is more deadly , Namely use append When , Will lock the whole watch , Other users even insert Other data will also be blocked .

Therefore, the production environment must not be used append,append It is also generally used for data extraction .

In fact, most of the time , use append Not much efficiency .

because append The reason why it is fast , It is because the log generation is reduced .

Only the following scenarios append It will reduce log generation ：

1、 In non Archive Mode

2、 In archive mode , The status of the table is nologging

First of all, non archive status is generally impossible , Slightly more important systems must be archived .

In archive mode nologging It's not usually , No log , This means that the recovered data will be incomplete in the future , It's risky .

Index Use

Use index Directly match the index to query data and improve the query speed

/+INDEX(TABLE INDEX_NAME)/

select /*+ index(x idx_t) */ * from t x where x.object_id=123

leading and use_nl Use

Use leading and use_nl To set the query order of the table , To speed up the query , Generally, the small table is set as the first table .

/+LEADING(TABLE)/

Take the specified table as the first table in the join order .

/+USE_NL(TABLE)/

Connect the specified table to the nested connected row source , And use the specified table as the internal table .

When making a multi table join query , [ The driver table ] For the definition of ：

1) When filter conditions are specified , The table with fewer rows of records that meet the query criteria is **[ The driver table ]**

2) When no filter condition is specified , The table with fewer rows is The driver table

Why Small tables drive large tables ？？

Generally speaking , No matter Oracle still Mysql, The goal of optimization is to reduce correlation as much as possible cycles , Ensure that the small table drives the large table

example : user surface 10000 Data ,class surface 20 Data

select * from user u left join class c u.userid=c.userid

This requires user Table cycle 10000 Times to find , And if used class Table driven user Tables only need to cycle 20 You can find it once

analysis ： No matter whether the large table is the driving table or the small table is the driving table , The number of comparisons is always 10000*20 Time ？？？

Advantages of small table driving large table

Advantage lies in ：

1. Large tables have indexes ： The time to query the large table is O(Log n)

1. A full scan of a large watch ： Fast disk block query

3 Exists On in The principle of big table driving small table in

Cost calculation method ：

Set a small table 100 That's ok , The big table 100000 That's ok .

Both tables have indexes ：

If the small watch is included , Big watch out ( The driver table ) Words , Then the number of scans is ：

100000+100000*2 ( among 2 Express IO frequency , An index , One time data )

If the big watch is included , The small watch is outside ( The driver table ) Words , Then the number of scans is ：

100+100*2.

Both tables have no index ：

If the small watch is included , If the big watch is outside , Then the number of scans is ：

100000+100*100000

If the big watch is included , If the watch is outside , Then the number of scans is ：

100+100000*100

Be careful ： If a table has an index , A table has no index ,ORACLE The table without index will be used as the driving table . If both tables have indexes , The outer surface is used as the driving table . If both are not indexed , Is also the appearance of the driving table .

Parallel Use

One 、 Parallel queries

Parallel queries allow a sql select The statement is divided into several smaller queries , Each part of the query runs concurrently , Then combine the results of each part , Provide the final result , Mostly used for full table scanning , Index full scan, etc , Scanning and linking of large tables 、 Create a large index 、 Partition index scan 、 Mass insert updates and deletions .

Enable parallel queries

SQL> ALTER TABLE T1 PARALLEL;

inform oracle, Yes T1 Enable parallel Inquire about , But the parallelism should be determined by referring to the resource load of the system .
utilize hints Tips , Enable parallel , At the same time, it can also tell the clear degree of parallelism , otherwise oracle Decide the parallelism you want to enable , These tips are only for this sql Statement is valid .

SQL> select /+ parallel(t1 8) / count(*) from t1;


SQL> select degree from user_tables where table_name='T1';

DEGREE

DEFAULT

The degree of parallelism is Default, Its value is given by 2 A parameter determines

SQL> show parameter cpu

NAME                                TYPE       VALUE

cpu_count                           integer    2
parallel_threads_per_cpu            integer    2

cpu_count Express cpu Count
parallel_threads_per_cpu Represent each cpu Number of parallel processes allowed
default Under the circumstances , The number of parallels is cpu_count*parallel_threads_per_cpu

Cancel the parallel setting

SQL> alter table t1 noparallel;

SQL> select degree from user_tables wheretable_name='T1';
DEGREE
1

Data dictionary view

v$px_session

sid： For each parallel session sid
qcsid：query coordinator sid, Query coordinator sid

Two 、 parallel dml

parallel dml Include insert,update,delete,merge, stay dml period ,oracle Multiple parallel execution servers can be used to execute insert,update,delete,merge, Multiple sessions execute simultaneously , At the same time, every session ( Concurrent processes ) All have their own undo paragraph , Are independent transactions , These matters are either handled by pdml Coordinator process commit , Or both rollback.

In one there is plenty I/o More bandwidth cpu In the host , For large scale dml, The speed may be greatly improved , Especially in large-scale data warehouse environment .

parallel dml The activation that needs to be displayed

SQL> alter session enable parallel dml;

Disable parallel dml

SQL> alter session disable parallel dml;

3、 ... and 、 parallel ddl

parallel ddl Provides dba The ability to use all machine resources , frequently-used ddl Yes
create table as select ……
create index
alter index rebuild
alter table move
alter table split

In these sql Add... After the statement parallel An example of a clause

SQL> alter table t1 move parallel;

Table altered

SQL> create index T1_IDX on T1 (OWNER,OBJECT_TYPE) tablespace SYSTEM parallel;

purpose

Force parallelism to execute the current SQL. This is in Oracle 9i Later versions can use , The previous version now has no environment for testing . in other words , Add this description , You can forcibly enable Oracle Multithreading function of . For example , It's like a multi-core computer CPU, But in most cases, it will not be fully enabled at the same time （2 More obvious above the nucleus ）, Use parallel explain , Multiple cores will work at the same time , To improve efficiency .

But start this function by itself , It also consumes resources and performance . all , Generally, the number of records returned is greater than 100 Ten thousand hour use , The effect will also be obvious .

grammar

/+parallel(table_short_name,cash_number)/

This can be added to insert、delete、update、select To use （ and rule It's almost the same , Have a chance to share rule Usage of ）

Turn on parallel The function statement is ：

alter session enable parallel dml;

This sentence is DML Sentences , If used in the program , use execute Method on .

Parallel It can also be used for multiple tables

If there are many tables , It's behind the first , Just add others . The specific writing method is as follows ：

/+parallel(t,10) (b,10)/

Summary

About execution efficiency , It is suggested to follow index Methods to improve the effect .Oracle It has its own explan road Methods , Before execution , Take a look at the implementation plan route first , Yes, written SQL tuned We'll do it later . There is no way out , Reuse parallel Method .Parallel More evil , For developers , It's not a good thing , Will form bad habits , Lead to many bad SQL It won't leak ,SQL Tuning Your ability cannot be improved . I have met some people create table after , never create index or primary key, Think of writing SQL Time plus parallel That's all right. .

There is something wrong , Welcome to the discussion .
Last , Welcome to pay attention to my wechat , What do you like , Collection , Forwarding is my greatest encouragement .