Teradata 字符串/文本搜索

Question

I have 2 tables in teradata. TableA and TableB

TableA has 2 columns as shown below

Table A

Column 1 --> Databasename

Column 2 --> Tablename

TableB has multiple columns including a text column.

I want to search databasename.tablename from Table A in Table B's Text column. The like operator cannot be used as there are around 2000 distinct table names in Table A. I have tried position join to do this as shown below but the query is extremely long running with high PJI and i had to manually abort it

select distinct a.Tablename ,b.text
from TableA a
inner join TableB b
on position(Trim(b.Text) in Trim('a.Databasename.'||a.tablename))>0
where b.theDate between add_months(date,-6) and date

UNION ALL

select distinct a.Tablename ,b.text
from TableA a
inner join TableB b
on position (Trim('a.Databasename.'||a.tablename) in Trim(b.Text))  >0
where b.theDate between add_months(date,-6) and date;

Is there an alternate way to do the above string search. Kindly share the SQL.

Thanks

Answer 1

REGEXP_SIMILAR:

One option is to use REGEXP_SIMILAR() which will be more precise than LIKE . I'm not sure if will be quicker, but it's worth a shot:

CREATE MULTISET VOLATILE TABLE TABLEA 
(databasename varchar(30), tablename varchar(30)) 
PRIMARY INDEX (databasename, tablename) ON COMMIT PRESERVE ROWS;

INSERT INTO TABLEA VALUES ('dba','tbla');
INSERT INTO TABLEA VALUES ('dba','tblb');
INSERT INTO TABLEA VALUES ('dbb','tbla');

CREATE MULTISET VOLATILE TABLE TABLEB 
(id int, sqlqry VARCHAR(5000)) 
ON COMMIT PRESERVE ROWS;

INSERT INTO TABLEB VALUES (1, 'SELECT * FROM dba.tbla;');
INSERT INTO TABLEB VALUES (2, 'SELECT smoecolumn FROM dba.tblb INNER JOIN dba.tbla ON foo = bar WHERe 1=1;');
INSERT INTO TABLEB VALUES (3, 'SELECT * FROM dbb.tbla WHERE foo=bar');

SELECT *
FROM TABLEA
    INNER JOIN TABLEB
        ON REGEXP_SIMILAR(TABLEB.sqlqry, '^.*' || TABLEA.databasename || '\.' || TABLEA.tablename || '.*$', 'i') = 1;

+-----+------+---+-----------------------------------------------------------------------------+
| dbb | tbla | 3 | SELECT * FROM dbb.tbla WHERE foo=bar                                        |
| dba | tbla | 2 | SELECT smoecolumn FROM dba.tblb INNER JOIN dba.tbla ON foo = bar WHERe 1=1; |
| dba | tbla | 1 | SELECT * FROM dba.tbla;                                                     |
| dba | tblb | 2 | SELECT smoecolumn FROM dba.tblb INNER JOIN dba.tbla ON foo = bar WHERe 1=1; |
+-----+------+---+-----------------------------------------------------------------------------+

---

STRTOK_SPLIT_TO_TABLE:

Here's where I was aiming with that strtok_split_to_tables comment. Basically you split your sql in TABLEB into words (splitting by space and ; characters). That will generate a row for every word.

From that list you just keep words that contain a period in them (like databasename.tablename ).

Then you can do your join on that between TABLEB and TABLEA:

CREATE MULTISET VOLATILE TABLE TABLEA 
(databasename varchar(30), tablename varchar(30)) 
PRIMARY INDEX (databasename, tablename) ON COMMIT PRESERVE ROWS;

INSERT INTO TABLEA VALUES ('dba','tbla');
INSERT INTO TABLEA VALUES ('dba','tblb');
INSERT INTO TABLEA VALUES ('dbb','tbla');

CREATE MULTISET VOLATILE TABLE TABLEB 
(id int, sqlqry VARCHAR(5000)) 
ON COMMIT PRESERVE ROWS;

INSERT INTO TABLEB VALUES (1, 'SELECT * FROM dba.tbla;');
INSERT INTO TABLEB VALUES (2, 'SELECT smoecolumn FROM dba.tblb INNER JOIN dba.tbla ON foo = bar WHERe 1=1;');
INSERT INTO TABLEB VALUES (3, 'SELECT * FROM dbb.tbla WHERE foo=bar');

WITH sqlwords AS
(
    SELECT tablebid, sqlwordnum, sqlword
    FROM TABLE (strtok_split_to_table(TABLEB.id, TABLEB.sqlqry, ' ;')
    RETURNS (tablebid integer, sqlwordnum integer, sqlword varchar(100)character set unicode) ) as sqlwordsplitter
    WHERE sqlwordsplitter.sqlword like '%.%'
)
SELECT TABLEA.*, TABLEB.*
FROM TABLEA
    INNER JOIN sqlwords
        ON TABLEA.databasename = strtok(sqlwords.sqlword, '.', 1)
            AND TABLEA.tablename = strtok(sqlwords.sqlword, '.', 2)
    INNER JOIN TABLEB
        ON sqlwords.tablebid = TABLEB.id;


+-----+------+---+-----------------------------------------------------------------------------+
| dbb | tbla | 3 | SELECT * FROM dbb.tbla WHERE foo=bar                                        |
| dba | tbla | 2 | SELECT smoecolumn FROM dba.tblb INNER JOIN dba.tbla ON foo = bar WHERe 1=1; |
| dba | tbla | 1 | SELECT * FROM dba.tbla;                                                     |
| dba | tblb | 2 | SELECT smoecolumn FROM dba.tblb INNER JOIN dba.tbla ON foo = bar WHERe 1=1; |
+-----+------+---+-----------------------------------------------------------------------------+

This isn't going to be super fast since we have to do word splitting, but it will definitely get the job done.

Answer 2

If it's for extracting a single tablename from CREATE TABLE AS you can apply Regular Expressions for table/DB name:

RegExp_Substr(SqlTextInfo, 'AS\s+?(.*?[.])?\K.+?\s+?(?=WITH\s)',1,1,'i') AS TableName
RegExp_Substr(SqlTextInfo, 'AS\s+?\K.*?(?=[.](.+?\s+)?WITH\s)',1,1,'i') AS DatabaseName

If the database name is missing you can COALESCE QryLogV.DefaultDatabase