Removing duplicates with unique index
Instead of IGNORE you can use ON DUPLICATE KEY UPDATE, which will give you control over which values should prevail.
If you have duplicates in your table and you use
ALTER TABLE mytable ADD UNIQUE INDEX myindex (A, B, C, D);
the query will fail with Error 1062 (duplicate key).
But if you use IGNORE
-- (only works before MySQL 5.7.4)
ALTER IGNORE TABLE mytable ADD UNIQUE INDEX myindex (A, B, C, D);
the duplicates will be removed. But the documentation doesn't specify which row will be kept:
IGNORE
is a MySQL extension to standard SQL. It controls howALTER TABLE
works if there are duplicates on unique keys in the new table or if warnings occur when strict mode is enabled. IfIGNORE
is not specified, the copy is aborted and rolled back if duplicate-key errors occur. IfIGNORE
is specified, only one row is used of rows with duplicates on a unique key. The other conflicting rows are deleted. Incorrect values are truncated to the closest matching acceptable value.
As of MySQL 5.7.4, the IGNORE clause for ALTER TABLE is removed and its use produces an error.
(ALTER TABLE Syntax)
If your version is 5.7.4 or greater - you can:
- Copy the data into a temporary table (it doesn't technically need to be temporary).
- Truncate the original table.
- Create the UNIQUE INDEX.
- And copy the data back with
INSERT IGNORE
(which is still available).
CREATE TABLE tmp_data SELECT * FROM mytable;
TRUNCATE TABLE mytable;
ALTER TABLE mytable ADD UNIQUE INDEX myindex (A, B, C, D);
INSERT IGNORE INTO mytable SELECT * from tmp_data;
DROP TABLE tmp_data;
If you use the
IGNORE
modifier, errors that occur while executing theINSERT
statement are ignored. For example, withoutIGNORE
, a row that duplicates an existingUNIQUE
index orPRIMARY KEY
value in the table causes a duplicate-key error and the statement is aborted. WithIGNORE
, the row is discarded and no error occurs. Ignored errors generate warnings instead.
(INSERT Syntax)
Also see: INSERT ... SELECT Syntax and Comparison of the IGNORE Keyword and Strict SQL Mode
if you think there will be duplicates, adding the unique index will fail. first check what duplicates there are:
select * from
(select a,b,c,d,count(*) as n from table_name group by a,b,c,d) x
where x.n > 1
This may be a expensive query on 20M rows, but will get you all duplicate keys that will prevent you from adding the primary index.
You could split this up into smaller chunks if you do a where in the subquery: where a='some_value'
For the records retrieved, you will have to change something to make the rows unique. If that is done (query returns 0 rows) you should be safe to add the primary index.