MySQL Triggers with Logging

By Michael McLaughlin, Planet MySQL – April 26, 2013 at 01:01AM

Somebody asked why you can’t implement MySQL triggers that write information when you want to stop the DML statement, like autonomous procedures in Oracle. The question was a surprise but I didn’t find anything on it, so here’s how you can do it. This is more or less like an autonomous process by leveraging both the InnoDB and MyISAM engine’s behaviors. This post leverages an earlier explanation of MySQL Triggers.

First you create a MyISAM table, which is a persistent store that auto commits when you’re other InnoDB tables can be transactionally dependent. Here’s a simple MyISAM logger table.

CREATE TABLE logger
( logger_id         INT UNSIGNED AUTO_INCREMENT PRIMARY KEY
, logger_event      VARCHAR(50)
, logger_table      VARCHAR(50)
, logger_instring   VARCHAR(100)
, logger_outstring  VARCHAR(100)
, created_by        INT UNSIGNED
, creation_date     DATE
, last_updated_by   INT UNSIGNED
, last_update_date  DATE) ENGINE=MyISAM;

Next, you create an on-insert trigger that changes an input but doesn’t stop the transaction. It also writes to the logger MyISAM table in the scope of the transaction.

CREATE TRIGGER contact_insert
BEFORE INSERT ON contact
FOR EACH ROW
BEGIN
 
  /* Check if last name contains a white space. */
  IF new.last_name REGEXP '^.* .*$' THEN
 
    /* Insert into an MyISAM table, which auto commits in the scope
       of a transaction. */
    INSERT INTO logger
    VALUES ( null
           ,'insert'
           ,'contact'
           , new.last_name
           , REPLACE(new.last_name,' ','-')
           , new.created_by
           , new.creation_date
           , new.last_updated_by
           , new.last_update_date );
 
    /* Replace the name for the INSERT INTO the CONTACT table. */
    SET new.last_name := REPLACE(new.last_name,' ','-');
  END IF;
END;
$$

Next, you create an on-update trigger that changes an update while aborting the transaction. It also writes to the logger MyISAM table because its outside the InnoDB scope of a transaction and auto committed on insert.

CREATE TRIGGER contact_update
BEFORE UPDATE ON contact
FOR EACH ROW
BEGIN
 
  /* Check if last name contains a white space. */
  IF new.last_name REGEXP '^.* .*$' THEN
 
    /* Insert into an MyISAM table, which auto commits in the scope
       of a transaction. */
    INSERT INTO logger
    VALUES ( null
           ,'update'
           ,'contact'
           , new.last_name
           , null           
           , old.created_by
           , old.creation_date
           , new.last_updated_by
           , new.last_update_date );
 
    /* Throw an exception to force the business user to see they
       can't update a last name with a white space. */
    SIGNAL SQLSTATE '42000';
  END IF;
END;
$$

Next, you create a test case with an INSERT and UPDATE statement that meets the condition of the triggers.

/* Insert a row meeting the trigger condition. */
INSERT INTO contact VALUES
( null, 1001, 1003,'Catherine', null,'Zeta Jones', 1001, UTC_DATE(), 1001, UTC_DATE());
 
/* Update a row meeting the trigger condition. */
UPDATE contact
SET    last_name = 'Zeta Jones'
,      last_updated_by = 1003
,      last_update_date = UTC_DATE()
WHERE  last_name = 'Zeta-Jones';

Last, query the logger table. You have a record inserted for both the allowed behavior and the aborted behavior. This means you have the ability to capture material that should never be inserted or updated into a table and who did it by leveraging the who-audit columns of the table.

SELECT * FROM logger;

It returns:

+-----------+--------------+--------------+-----------------+------------------+------------+---------------+-----------------+------------------+
| logger_id | logger_event | logger_table | logger_instring | logger_outstring | created_by | creation_date | last_updated_by | last_update_date |
+-----------+--------------+--------------+-----------------+------------------+------------+---------------+-----------------+------------------+
|         1 | insert       | contact      | Zeta Jones      | Zeta-Jones       |       1001 | 2013-04-26    |            1001 | 2013-04-26       |
|         2 | update       | contact      | Zeta Jones      | NULL             |       1001 | 2013-04-26    |            1003 | 2013-04-26       |
+-----------+--------------+--------------+-----------------+------------------+------------+---------------+-----------------+------------------+
2 rows in set (0.00 sec)

This effectively delivers in MySQL the equivalent of an autonomous transaction in Oracle. The result from the non-critical trigger records the before and after value, while the results from the critical update trigger only record the before values because the event is aborted by raising an error in the trigger. As always, I hope this helps somebody looking for a solution.

PlanetMySQL Voting:
Vote UP /
Vote DOWN

Share this: