The first option is to upgrade to MySQL version 4.1.

The second option is to use a procedural programming language (such as Perl or PHP) to submit a SELECT query to obtain the primary keys for the records to be deleted, and then use these values to construct the DELETE statement (DELETE FROM ... WHERE ... IN (key1, key2, ...)).

The third option is to use interactive SQL to construct a set of DELETE statements automatically, using the MySQL extension CONCAT() (in lieu of the standard || operator). For example:

SELECT CONCAT('DELETE FROM tab1 WHERE pkid = ', "'", tab1.pkid, "'", ';')
  FROM tab1, tab2
 WHERE tab1.col1 = tab2.col2;

You can place this query in a script file and redirect input from it to the mysql command-line interpreter, piping its output back to a second instance of the interpreter:

shell mysql --skip-column-names mydb  myscript.sql | mysql mydb

Loops that need transactions normally can be coded with the help of LOCK TABLES, and you don't need cursors when you can update records on the fly.

To avoid using ROLLBACK, you can use the following strategy:

This is usually a much faster method than using transactions with possible ROLLBACKs, although not always. The only situation this solution doesn't handle is when someone kills the threads in the middle of an update. In this case, all locks will be released but some of the updates may not have been executed.

You can also use functions to update records in a single operation. You can get a very efficient application by using the following techniques:

For example, when we are doing updates to some customer information, we update only the customer data that has changed and test only that none of the changed data, or data that depends on the changed data, has changed compared to the original row. The test for changed data is done with the WHERE clause in the UPDATE statement. If the record wasn't updated, we give the client a message: "Some of the data you have changed has been changed by another user." Then we show the old row versus the new row in a window, so the user can decide which version of the customer record he should use.

This gives us something that is similar to column locking but is actually even better because we only update some of the columns, using values that are relative to their current values. This means that typical UPDATE statements look something like these:

UPDATE tablename SET pay_back=pay_back+125;

UPDATE customer
  SET
    customer_date='current_date',
    address='new address',
    phone='new phone',
    money_he_owes_us=money_he_owes_us-125
  WHERE
    customer_id=id AND address='old address' AND phone='old phone';

As you can see, this is very efficient and works even if another client has changed the values in the pay_back or money_he_owes_us columns.

In many cases, users have wanted ROLLBACK and/or LOCK TABLES for the purpose of managing unique identifiers for some tables. This can be handled much more efficiently by using an AUTO_INCREMENT column and either the SQL function LAST_INSERT_ID() or the C API function mysql_insert_id(). Section 9.1.3.31, “mysql_insert_id()”.

You can generally code around row-level locking. Some situations really need it, but they are very few. InnoDB tables support row-level locking. With MyISAM, you can use a flag column in the table and do something like the following:

UPDATE tbl_name SET row_flag=1 WHERE id=ID;

MySQL returns 1 for the number of affected rows if the row was found and row_flag wasn't already 1 in the original row.

You can think of it as though MySQL Server changed the preceding query to:

UPDATE tbl_name SET row_flag=1 WHERE id=ID AND row_flag  1;

Assuming proper design of the relations, foreign key constraints will make it more difficult for a programmer to introduce an inconsistency into the database.

Using cascading updates and deletes can simplify the client code.

Properly designed foreign key rules aid in documenting relations between tables.

Mistakes, which are easy to make in designing key relations, can cause severe problems--for example, circular rules, or the wrong combination of cascading deletes.

A properly written application will make sure (internally) that it is not violating referential integrity constraints before proceding with a query. Thus, additional checks on the database level will only slow down performance for such an application.

It is not uncommon for a DBA to make such a complex topology of relations that it becomes very difficult, and in some cases impossible, to back up or restore individual tables.

One can get a deadlock when doing LOCK TABLE on multiple tables and then in the same connection doing a DROP TABLE on one of them while another thread is trying to lock the table. One can however do a KILL on any of the involved threads to resolve this. Fixed in 4.0.12.

SELECT MAX(key_column) FROM t1,t2,t3... where one of the tables are empty doesn't return NULL but instead the maximum value for the column. Fixed in 4.0.11.

DELETE FROM heap_table without a WHERE doesn't work on a locked HEAP table.

FLUSH TABLES WITH READ LOCK does not block CREATE TABLE or COMMIT, which make cause a problem with the binary log position when doing a full backup of tables and the binary log.

ANALYZE TABLE on a BDB table may in some cases make the table unusable until one has restarted mysqld. When this happens you will see errors like the following in the MySQL error file:

001207 22:07:56  bdb:  log_flush: LSN past current end-of-log

MySQL accepts parentheses in the FROM part, but silently ignores them. The reason for not giving an error is that many clients that automatically generate queries add parentheses in the FROM part even where they are not needed.

Concatenating many RIGHT JOINS or combining LEFT and RIGHT join in the same query may not give a correct answer as MySQL only generates NULL rows for the table preceding a LEFT or before a RIGHT join. This will be fixed in 5.0 at the same time we add support for parentheses in the FROM part.

Don't execute ALTER TABLE on a BDB table on which you are running multi-statement transactions until all those transactions complete. (The transaction will probably be ignored.)

ANALYZE TABLE, OPTIMIZE TABLE, and REPAIR TABLE may cause problems on tables for which you are using INSERT DELAYED.

Doing a LOCK TABLE ... and FLUSH TABLES ... doesn't guarantee that there isn't a half-finished transaction in progress on the table.

BDB tables are a bit slow to open. If you have many BDB tables in a database, it will take a long time to use the mysql client on the database if you are not using the -A option or if you are using rehash. This is especially notable when you have a big table cache.

Replication uses query-level logging: the master writes the executed queries to the binary log. This is a very fast, compact and efficient logging method which works perfectly in most cases. However, currently there is a theoretical chance (though we never heard about it coming true) that the data on the master and slave become different if a query is designed in such a way that the data modification is non-deterministic, that is, left to the will of the query optimiser (which generally is no good practice, even outside of replication!). For example:

IF and only if all these queries have NO ORDER BY clause guaranteeing a deterministic order.

Indeed, for example for INSERT ... SELECT with no ORDER BY, the SELECT may return rows in a different order (which will result in a row having different ranks, hence getting a different number in the auto_increment column), depending on the choices made by the optimisers on the master and slave. A query will be optimised differently on the master and slave only if :

This problem may also affect database restoration using mysqlbinlog|mysql.

The easiest way to avoid this problem in all cases is add an ORDER BY clause to such non-deterministic queries to ensure that the rows are always stored/modified in the same order. In future MySQL versions we will automatically add an ORDER BY clause when needed.

LIKE is not multi-byte character safe. Comparison is done character by character.

When using RPAD function, or any other string function that ends up adding blanks to the right, in a query that has to use temporary table to be resolved, then all resulting strings will be RTRIM'ed. This is an example of the query:

SELECT RPAD(t1.field1, 50, ' ') AS f2, RPAD(t2.field2, 50, ' ') AS f1 FROM table1 as t1 LEFT JOIN table2 AS t2 ON t1.record=t2.joinID ORDER BY t2.record;

Final result of this bug is that use will not be able to get blanks on the right side of the resulting field.

The above behaviour exists in all versions of MySQL.

The reason for this is due to the fact that HEAP tables, which are used first for temporary tables, are not capable of handling VARCHAR columns.

This behaviour will be fixed in one of the 4.1 series releases.

Because of how table definitions files are stored one can't use character 255 (CHAR(255)) in table names, column names or enums. This is scheduled to be fixed in version 5.1 when we have new table definition format files.

When using SET CHARACTER SET, one can't use translated characters in database, table, and column names.

One can't use _ or % with ESCAPE in LIKE ... ESCAPE.

If you have a DECIMAL column with a number stored in different formats (+01.00, 1.00, 01.00), GROUP BY may regard each value as a different value.

DELETE FROM merge_table used without a WHERE will only clear the mapping for the table, not delete everything in the mapped tables.

You cannot build the server in another directory when using MIT-pthreads. Because this requires changes to MIT-pthreads, we are not likely to fix this. Section 2.3.6, “MIT-pthreads Notes ”.

BLOB values can't "reliably" be used in GROUP BY or ORDER BY or DISTINCT. Only the first max_sort_length bytes (default 1024) are used when comparing BLOBs in these cases. This can be changed with the -O max_sort_length option to mysqld. A workaround for most cases is to use a substring: SELECT DISTINCT LEFT(blob,2048) FROM tbl_name.

Calculation is done with BIGINT or DOUBLE (both are normally 64 bits long). It depends on the function which precision one gets. The general rule is that bit functions are done with BIGINT precision, IF, and ELT() with BIGINT or DOUBLE precision and the rest with DOUBLE precision. One should try to avoid using unsigned long long values if they resolve to be bigger than 63 bits (9223372036854775807) for anything else than bit fields. MySQL Server 4.0 has better BIGINT handling than 3.23.

All string columns, except BLOB and TEXT columns, automatically have all trailing spaces removed when retrieved. For CHAR types this is okay, and may be regarded as a feature according to SQL-92. The bug is that in MySQL Server, VARCHAR columns are treated the same way.

You can only have up to 255 ENUM and SET columns in one table.

In MIN(), MAX() and other aggregate functions, MySQL currently compares ENUM and SET columns by their string value rather than by the string's relative position in the set.

mysqld_safe redirects all messages from mysqld to the mysqld log. One problem with this is that if you execute mysqladmin refresh to close and reopen the log, stdout and stderr are still redirected to the old log. If you use -log extensively, you should edit mysqld_safe to log to 'hostname'.err instead of 'hostname'.log so you can easily reclaim the space for the old log by deleting the old one and executing mysqladmin refresh.

In the UPDATE statement, columns are updated from left to right. If you refer to an updated column, you will get the updated value instead of the original value. For example:

mysql UPDATE tbl_name SET KEY=KEY+1,KEY=KEY+1;

This will update KEY with 2 instead of with 1.

You can refer to multiple temporary tables in the same query, but you cannot refer to any given temporary table more than once. For example, the following doesn't work:

mysql SELECT * FROM temporary_table, temporary_table AS t2;

RENAME doesn't work with TEMPORARY tables or tables used in a MERGE table.

The optimiser may handle DISTINCT differently if you are using 'hidden' columns in a join or not. In a join, hidden columns are counted as part of the result (even if they are not shown) while in normal queries hidden columns don't participate in the DISTINCT comparison. We will probably change this in the future to never compare the hidden columns when executing DISTINCT.

An example of this is:

SELECT DISTINCT mp3id FROM band_downloads
       WHERE userid = 9 ORDER BY id DESC;

and

SELECT DISTINCT band_downloads.mp3id
       FROM band_downloads,band_mp3
       WHERE band_downloads.userid = 9
       AND band_mp3.id = band_downloads.mp3id
       ORDER BY band_downloads.id DESC;

In the second case you may in MySQL Server 3.23.x get two identical rows in the result set (because the hidden id column may differ).

Note that this happens only for queries where you don't have the ORDER BY columns in the result, something that you are not allowed to do in SQL-92.

Because MySQL Server allows you to work with table types that don't support transactions, and thus can't rollback data, some things behave a little differently in MySQL Server than in other SQL servers. This is just to ensure that MySQL Server never needs to do a rollback for an SQL command. This may be a little awkward at times as column values must be checked in the application, but this will actually give you a nice speed increase as it allows MySQL Server to do some optimisations that otherwise would be very hard to do.

If you set a column to an incorrect value, MySQL Server will, instead of doing a rollback, store the best possible value in the column:

If you execute a PROCEDURE on a query that returns an empty set, in some cases the PROCEDURE will not transform the columns.

Creation of a table of type MERGE doesn't check if the underlying tables are of compatible types.

MySQL Server can't yet handle NaN, -Inf, and Inf values in double. Using these will cause problems when trying to export and import data. We should as an intermediate solution change NaN to NULL (if possible) and -Inf and Inf to the minimum respective maximum possible double value.

LIMIT on negative numbers are treated as big positive numbers.

If you use ALTER TABLE to first add a UNIQUE index to a table used in a MERGE table and then use ALTER TABLE to add a normal index on the MERGE table, the key order will be different for the tables if there was an old key that was not unique in the table. This is because ALTER TABLE puts UNIQUE keys before normal keys to be able to detect duplicate keys as early as possible.

You can get a hung thread if you do a DROP TABLE on a table that is one among many tables that is locked with LOCK TABLES.

In the following case you can get a core dump:

Before MySQL Server Version 3.23.2 an UPDATE that updated a key with a WHERE on the same key may have failed because the key was used to search for records and the same row may have been found multiple times:

UPDATE tbl_name SET KEY=KEY+1 WHERE KEY  100;

A workaround is to use:

mysql UPDATE tbl_name SET KEY=KEY+1 WHERE KEY+0  100;

This will work because MySQL Server will not use an index on expressions in the WHERE clause.

Before MySQL Server Version 3.23, all numeric types were treated as fixed-point fields. That means you had to specify how many decimals a floating-point field shall have. All results were returned with the correct number of decimals.