MacLochlainns Weblog

Michael McLaughlin's Technical Blog

Site Admin

Archive for the ‘MyISAM’ Category

MySQL Triggers with Logging

with one comment

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.

  1. 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;
  1. 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;
$$

  1. 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;
$$

  1. 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';

  1. 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.

Written by maclochlainn

April 25th, 2013 at 11:01 pm

Posted in InnoDB,MyISAM,MySQL,MySQL Trigger

Tagged with , , , ,

Delay or synchronize it?

with one comment

A couple students in one of my classes ran into a problem when competing Java threads tried to insert new rows in a table. They raised an error when they tried the DELAY keyword to avoid the race (collision) condition in an INSERT statement. It was simple to explain to them that the DELAY keyword doesn’t work with an InnoDB table. Any attempt throws the following error:

ERROR 1616 (HY000): DELAYED OPTION NOT supported FOR TABLE 'message'

Important Update: INSERT DELAYED is gone in MySQL 5.6.6 (announcement) and the whole issue comes down to synchronizing threads (some dislike the solution) or using the ON DUPLICATE KEY clause.

They retested their Java application after redefining the target table using the MyISAM engine. They found it worked but that’s a bad fix in Java (a brief Java/MySQL tutorial post). They really needed to synchronize the Java thread (line #22), leave out the DELAY keyword, and manage the table with the InnoDB engine. Here’s the modified Java code (by the way, they named their project VulcanTech if you’re wondering about the packages in the import statement):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33

package vulcantech.vth.server.commands;
 
import java.io.IOException;
import java.net.Socket;
 
import vulcantech.vth.server.combeans.MessageBean;
 
public class MessageHandler implements Handler {
 
  @Override
  public void handleIt(Object... args) {
 
    MessageBean message = (MessageBean) args[0];
    Socket sock = (Socket) args[1];
 
    DatabaseConnection dbconnection = new DatabaseConnection();
 
    String update = new String("INSERT INTO message(message_timestamp, sender, recipient, message, checked) VALUES(\'" 
                  + message.getTimeStamp() + "\', \'" + message.getSender() + "\', \'" 
                  + message.getRecipient() + "\', \'" + message.getMessage() + "\', b\'0\')");
 
    synchronized (this) {
      dbconnection.executeUpdate(update);
    }
 
    try {
      sock.getOutputStream().write(1);
    } catch (IOException e) {
      e.printStackTrace();
    }
    dbconnection.close();
  }
}

Hope this helps those who encounter race conditions against MySQL when you’re writing Enterprise Java Beans (EJBs).

Written by maclochlainn

July 22nd, 2012 at 3:53 pm

Posted in InnoDB,Java,Linux,Mac OS X,MyISAM,MySQL,Windows7

Tagged with , ,