SQL: How to find duplicates based on two fields?

Question

I have rows in an Oracle database table which should be unique for a combination of two fields but the unique constrain is not set up on the table so I need to find all rows which violate the constraint myself using SQL. Unfortunately my meager SQL skills aren't up to the task.

My table has three columns which are relevant: entity_id, station_id, and obs_year. For each row the combination of station_id and obs_year should be unique, and I want to find out if there are rows which violate this by flushing them out with an SQL query.

I have tried the following SQL (suggested by this previous question) but it doesn't work for me (I get ORA-00918 column ambiguously defined):

SELECT
entity_id, station_id, obs_year
FROM
mytable t1
INNER JOIN (
SELECT entity_id, station_id, obs_year FROM mytable 
GROUP BY entity_id, station_id, obs_year HAVING COUNT(*) > 1) dupes 
ON 
t1.station_id = dupes.station_id AND
t1.obs_year = dupes.obs_year

Can someone suggest what I'm doing wrong, and/or how to solve this?

Answer 1

You need to specify the table for the columns in the main select. Also, assuming entity_id is the unique key for mytable and is irrelevant to finding duplicates, you should not be grouping on it in the dupes subquery.

Try:

SELECT t1.entity_id, t1.station_id, t1.obs_year
FROM mytable t1
INNER JOIN (
SELECT station_id, obs_year FROM mytable 
GROUP BY station_id, obs_year HAVING COUNT(*) > 1) dupes 
ON 
t1.station_id = dupes.station_id AND
t1.obs_year = dupes.obs_year

Answer 2

SELECT  *
FROM    (
        SELECT  t.*, ROW_NUMBER() OVER (PARTITION BY station_id, obs_year ORDER BY entity_id) AS rn
        FROM    mytable t
        )
WHERE   rn > 1

Answer 3

Could you not create a new table that includes the unique constraint, and then copy across the data row by row, ignoring failures?

Answer 4

SELECT  *
FROM    (
        SELECT  t.*, ROW_NUMBER() OVER (PARTITION BY station_id, obs_year ORDER BY entity_id) AS rn
        FROM    mytable t
        )
WHERE   rn > 1

by Quassnoi is the most efficient for large tables. I had this analysis of cost :

SELECT a.dist_code, a.book_date, a.book_no
FROM trn_refil_book a
WHERE EXISTS (SELECT 1 from trn_refil_book b Where
       a.dist_code = b.dist_code and a.book_date = b.book_date and a.book_no = b.book_no
       AND a.RowId <> b.RowId)
       ;

gave a cost of 1322341

SELECT a.dist_code, a.book_date, a.book_no
FROM trn_refil_book a
INNER JOIN (
SELECT b.dist_code, b.book_date, b.book_no FROM trn_refil_book b 
GROUP BY b.dist_code, b.book_date, b.book_no HAVING COUNT(*) > 1) c 
ON 
 a.dist_code = c.dist_code and a.book_date = c.book_date and a.book_no = c.book_no
;

gave a cost of 1271699

while

SELECT  dist_code, book_date, book_no
FROM    (
        SELECT  t.dist_code, t.book_date, t.book_no, ROW_NUMBER() OVER (PARTITION BY t.book_date, t.book_no
          ORDER BY t.dist_code) AS rn
        FROM    trn_refil_book t
        ) p
WHERE   p.rn > 1
;

gave a cost of 1021984

The table was not indexed....

Answer 5

Change the 3 fields in the initial select to be

SELECT
t1.entity_id, t1.station_id, t1.obs_year

Answer 6

SELECT entity_id, station_id, obs_year
FROM mytable t1
WHERE EXISTS (SELECT 1 from mytable t2 Where
       t1.station_id = t2.station_id
       AND t1.obs_year = t2.obs_year
       AND t1.RowId <> t2.RowId)