MySQL insert on duplicate update for non-PRIMARY key

Question

I am little confused with insert on duplicate update query. I have MySQL table with structure like this:

• record_id (PRIMARY, UNIQUE)
• person_id (UNIQU

Answer 1

Your question is very valid. This is a very common requirement. And most people get it wrong, due to what MySQL offers.

• The requirement: Insert unless the PRIMARY key exists, otherwise update.
• The common approach: ON DUPLICATE KEY UPDATE
• The result of that approach, disturbingly: Insert unless the PRIMARY or any UNIQUE key exists, otherwise update!

What can go horribly wrong with ON DUPLICATE KEY UPDATE? You insert a supposedly new record, with a new PRIMARY key value (say a UUID), but you happen to have a duplicate value for its UNIQUE key.

What you want is a proper exception, indicating that you are trying to insert a duplicate into a UNIQUE column.

But what you get is an unwanted UPDATE! MySQL will take the conflicting record and start overwriting its values. If this happens unintentionally, you have mutilated an old record, and any incoming references to the old record are now referencing the new record. And since you probably won't tell the query to update the PRIMARY column, your new UUID is nowhere to be found. If you ever encounter this data, it will probably make no sense and you will have no idea where it came from.

We need a solution to actually insert unless the PRIMARY key exists, otherwise update.

We will use a query that consists of two statements:

1. Update where the PRIMARY key value matches (affects 0 or 1 rows).
2. Insert if the PRIMARY key value does not exist (inserts 1 or 0 rows).

This is the query:

UPDATE my_table SET
unique_name = 'one', update_datetime = NOW()
WHERE id = 1;

INSERT INTO my_table
SELECT 1, 'one', NOW()
FROM my_table
WHERE id = 1
HAVING COUNT(*) = 0;

Only one of these queries will have an effect. The UPDATE is easy. As for the INSERT: WHERE id = 1 results in a row if the id exists, or no row if it does not. HAVING COUNT(*) = 0 inverts that, resulting in a row if the id is new, or no row if it already exists.

I have explored other variants of the same idea, such as with a LEFT JOIN and WHERE, but they all looked more convoluted. Improvements are welcome.

Answer 2

How about my approach?

Let's say you have one table with a autoincrement id and three text-columns. You want to insert/update the value of column3 with the values in column1 and column2 being a (non unique) key.

I use this query (without explicitly locking the table):

insert into myTable (id, col1, col2, col3)
select tmp.id, 'col1data', 'col2data', 'col3data' from
(select id from myTable where col1 = 'col1data' and col2 = 'col2data' union select null as id limit 1) tmp
on duplicate key update col3 = values(col3)

Anything wrong with that? For me it works the way I want.

Answer 3

13.2.5.3 INSERT ... ON DUPLICATE KEY UPDATE Syntax

If you specify ON DUPLICATE KEY UPDATE, and a row is inserted that would cause a duplicate value in a UNIQUE index or PRIMARY KEY, MySQL performs an UPDATE of the old row.

Example:

DELIMITER //

DROP PROCEDURE IF EXISTS `sp_upsert`//
DROP TABLE IF EXISTS `table_test`//

CREATE TABLE `table_test` (
  `record_id` INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,
  `person_id` INT UNSIGNED NOT NULL,
  `some_text` VARCHAR(50),
  `some_other_text` VARCHAR(50),
  UNIQUE KEY `record_id_index` (`record_id`),
  UNIQUE KEY `person_id_index` (`person_id`)
)//

INSERT INTO `table_test`
  (`person_id`, `some_text`, `some_other_text`)
VALUES
  (1, 'AAA', 'XXX'),
  (2, 'BBB', 'YYY'),
  (3, 'CCC', 'ZZZ')//

CREATE PROCEDURE `sp_upsert`(
  `p_person_id` INT UNSIGNED,
  `p_some_text` VARCHAR(50),
  `p_some_other_text` VARCHAR(50)
)
BEGIN
  INSERT INTO `table_test`
    (`person_id`, `some_text`, `some_other_text`)
  VALUES
    (`p_person_id`, `p_some_text`, `p_some_other_text`)
  ON DUPLICATE KEY UPDATE `some_text` = `p_some_text`,
                          `some_other_text` = `p_some_other_text`;
END//

DELIMITER ;

mysql> CALL `sp_upsert`(1, 'update_text_0', 'update_text_1');
Query OK, 2 rows affected (0.00 sec)

mysql> SELECT
    ->   `record_id`,
    ->   `person_id`,
    ->   `some_text`,
    ->   `some_other_text`
    -> FROM
    ->   `table_test`;
+-----------+-----------+---------------+-----------------+
| record_id | person_id | some_text     | some_other_text |
+-----------+-----------+---------------+-----------------+
|         1 |         1 | update_text_0 | update_text_1   |
|         2 |         2 | BBB           | YYY             |
|         3 |         3 | CCC           | ZZZ             |
+-----------+-----------+---------------+-----------------+
3 rows in set (0.00 sec)

mysql> CALL `sp_upsert`(4, 'new_text_0', 'new_text_1');
Query OK, 1 row affected (0.00 sec)

mysql> SELECT
    ->   `record_id`,
    ->   `person_id`,
    ->   `some_text`,
    ->   `some_other_text`
    -> FROM
    ->   `table_test`;
+-----------+-----------+---------------+-----------------+
| record_id | person_id | some_text     | some_other_text |
+-----------+-----------+---------------+-----------------+
|         1 |         1 | update_text_0 | update_text_1   |
|         2 |         2 | BBB           | YYY             |
|         3 |         3 | CCC           | ZZZ             |
|         5 |         4 | new_text_0    | new_text_1      |
+-----------+-----------+---------------+-----------------+
4 rows in set (0.00 sec)

SQL Fiddle demo

Answer 4

You need a query that check if exists any row with you record_id (or person_id). If exists update it, else insert new row

IF EXISTS (SELECT * FROM table.person WHERE record_id='SomeValue')
    UPDATE table.person 
    SET some_text='new_some_text', some_other_text='some_other_text' 
    WHERE record_id='old_record_id'
ELSE
    INSERT INTO table.person (record_id, person_id, some_text, some_other_text) 
    VALUES ('new_record_id', 'new_person_id', 'new_some_text', 'new_some_other_text')

Another better approach is

UPDATE table.person SET (...) WHERE person_id='SomeValue'
IF ROW_COUNT()=0
    INSERT INTO table.person (...) VALUES (...)

Answer 5

I came across this post because I needed what's written in the title, and I found a pretty handy solution, but no one mentioned it here, so I thought of pasting it here. Note that this solution is very handy if you're initiating your database tables. In this case, when you create your corresponding table, define your primary key etc. as usual, and for the combination of columns you want to be unique, simply add

UNIQUE(column_name1,column_name2,...)

at the end of your CREATE TABLE statement, for any combination of the specified columns you want to be unique. Like this, according to this page here, "MySQL uses the combination of values in both column column_name1 and column_name2 to evaluate the uniqueness", and reports an error if you try to make an insert which already has the combination of values for column_name1 and column_name2 you provide in your insert. Combining this way of creating a database table with the corresponding INSERT ON DUPLICATE KEY syntax appeared to be the most suitable solution for me. Just need to think of it carefully before you actually start using your table; when setting up your database tables.

Answer 6

A flexible solution should retain the atomicity offered by INSERT ... ON DUPLICATE KEY UPDATE and work regardless of if it's autocommit=true and not depend on a transaction with an isolation level of REPEATABLE READ or greater.

Any solution performing check-then-act across multiple statements would not satisfy this.

Here are the options:

If there tends to be more inserts than updates:

INSERT INTO table (record_id, ..., some_text, some_other_text) VALUES (...);

IF <duplicate entry for primary key error>
  UPDATE table SET some_text = ..., some_other_text = ... WHERE record_id = ...;

  IF affected-rows = 0
    -- retry from INSERT OR ignore this conflict and defer to the other session

If there tends to be more updates than inserts:

UPDATE table SET some_text = ..., some_other_text = ... WHERE record_id = ...;

IF affected-rows = 0
  INSERT INTO table (record_id, ..., some_text, some_other_text) VALUES (...);

  IF <duplicate entry for primary key error>
    -- retry from UPDATE OR ignore this conflict and defer to the other session

If you don't mind a bit of ugliness, you can actually use INSERT ... ON DUPLICATE KEY UPDATE and do this in a single statement:

INSERT INTO table (record_id, ..., some_text, some_other_text) VALUES (...)
    ON DUPLICATE KEY UPDATE
      some_text = if(record_id = VALUES(record_id), VALUES(some_text), some_text),
      some_other_text = if(record_id = VALUES(record_id), VALUES(some_other_text), some_other_text)

IF affected-rows = 0
  -- handle this as a unique check constraint violation

Note: affected-rows in these examples mean affected rows and not found rows. The two can be confused because a single parameter switches which of these values the client is returned.

Also note, if some_text and some_other_text are not actually modified (and the record is not otherwise changed) when you perform the update, those checks on affected-rows = 0 will misfire.