LIKE statement taking too long

问题

Please see the DDL below:

create table #VRMs (ID INT, VRM VARCHAR(10))
INSERT INTO #VRMs VALUES (1,'VRM1')
INSERT INTO #VRMs VALUES (1,'VRM2')
INSERT INTO #VRMs VALUES (1,'VRM3')

The users want to be able to search for VRMs like this:

select * from #VRMs where VRM LIKE '%VRM1'

The problem is that there are 100,000,000 rows in this table and the queries are taking far too long. I have come across this question: https://stackoverflow.com/questions/33210269/like-statement-taking-too-long

Unfortunately I am using an Oracle DBMS. I have imported all the VRMs into an SQL database using SSIS and have tried the following query:

select * from #VRMs where reverse(VRM) LIKE reverse('%VRM1')

The query runs in a fraction of a second. However, it runs in 487 seconds on the Oracle database. How can I optimise it to run faster in the Oracle database?

回答1:

Use function based index

 create index tst_rev on tst(reverse(txt));

The query

 select * from tst where reverse(txt) like 'x10%';

 101x
 201x
 301x
 401x

leads to index range scan

 ---------------------------------------------------------------------------------------
 | Id  | Operation                   | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
 ---------------------------------------------------------------------------------------
 |   0 | SELECT STATEMENT            |         |   100 | 10400 |     0   (0)| 00:00:01 |
 |   1 |  TABLE ACCESS BY INDEX ROWID| TST     |   100 | 10400 |     0   (0)| 00:00:01 |
 |*  2 |   INDEX RANGE SCAN          | TST_REV |    18 |       |     0   (0)| 00:00:01 |
 ---------------------------------------------------------------------------------------

 ....

    2 - access(REVERSE("TXT") LIKE 'x10%')
   filter(REVERSE("TXT") LIKE 'x10%')

回答2:

your query takes seconds in MSSQL bcz you imported everything into temp table #VRM

in oracle you created #VRM table, the table is not temp table like it was in MSSQL, and then your imported millions of records there I think you need create index for the column VRM in the table #VRM

回答3:

Doing a quick search .. I found this link on AskTom ..

https://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:37336026927381

I think the solution posted by a user part way down could be useful .. To avoid any future broken links, I re-post the original in it's original form .. (with just some minor formatting editting ... ):

Why not use a builtin functionality?

March 27, 2005 - 3:40 am UTC

Reviewer: Ofir Manor from Israel

Hi Tom,

I wonder why you hand-code a functionality that is already supplied? If you want to make '%xxx%' queries faster, all you need to do is ask the text index to prepare for it, by indexing all the substrings as well. There is no need to manually write a procedural code to implement this feature... Did I miss anything?

begin ctx_ddl.create_preference('SUBSTRING_PREF', 'BASIC_WORDLIST'); ctx_ddl.set_attribute('SUBSTRING_PREF', 'SUBSTRING_INDEX','TRUE'); end; drop index search_idx; create index search_idx on T(col1) indextype is ctxsys.context parameters ('wordlist SUBSTRING_PREF MEMORY 50M'); ..

Then this query will do:

select * from t where contains( col1,'%750776%')>0

There are some more details and preference attributes to play with here, plus an example:

http://download-west.oracle.com/docs/cd/B14117_01/text.101/b10730/cdatadic.htm#sthref750

How about that?

来源：https://stackoverflow.com/questions/33302679/like-statement-taking-too-long