D.3.2 Changes in release 4.0.14 (18 Jul 2003)

Functionality added or changed:

• A documentation change: Function INTERVAL(NULL, ...) returns -1.
• Enabled INSERT from SELECT when the table into which the records are inserted is also a table listed in the SELECT.
• Allow CREATE TABLE and INSERT from any UNION.
• The SQL_CALC_FOUND_ROWS option now always returns the total number of rows for any UNION.
• Removed --table option from mysqlbinlog to avoid repeating mysqldump functionality.
• Changed optimiser slightly to prefer index lookups over full table scans in some boundary cases.
• Added thread-specific max_seeks_for_key variable that can be used to force the optimiser to use keys instead of table scans even if the cardinality of the index is low.
• Added optimisation that converts LEFT JOIN to normal join in some cases.
• A documentation change: added a paragraph about failover in replication (how to use a surviving slave as the new master, how to resume to the original setup). See section 4.10.8 Replication FAQ.
• A documentation change: added warning notes about safe use of the CHANGE MASTER command. See section 4.10.7.9 CHANGE MASTER TO master_def_list (slave).
• MySQL now issues a warning (not an error, as in 4.0.13) when it opens a table that was created with MySQL 4.1.
• Added --nice option to mysqld_safe to allow setting the niceness of the mysqld process. (Thanks to Christian Hammers for providing the initial patch.) (Bug #627)
• Added --read-only option to cause mysqld to allow no updates except from slave threads or from users with the SUPER privilege. (Original patch from Markus Benning).
• SHOW BINLOG EVENTS FROM x where x is strictly less than 4 now silently converts x to 4 instead of printing an error. The same change was done for CHANGE MASTER TO MASTER_LOG_POS=x and CHANGE MASTER TO RELAY_LOG_POS=x.
• mysqld now only adds an interrupt handler for the SIGINT signal if you start it with the new --gdb option. This is because some MySQL users encountered strange problems when they accidentally sent SIGINT to mysqld threads.
• RESET SLAVE now clears the Last_errno and Last_error fields in the output of SHOW SLAVE STATUS.
• Added max_relay_log_size variable; the relay log will be rotated automatically when its size exceeds max_relay_log_size. But if max_relay_log_size is 0 (the default), max_binlog_size will be used (as in older versions). max_binlog_size still applies to binary logs in any case.
• FLUSH LOGS now rotates relay logs in addition to the other types of logs it already rotated.

Bugs fixed:

• Comparison/sorting for latin1_de character set was rewritten. The old algorithm could not handle cases like "sä" > "ßa". See section 4.6.1.1 German character set. In rare cases it resulted in table corruption.
• Fixed a problem with the password prompt on Windows. (Bug #683)
• ALTER TABLE ... UNION=(...) for MERGE table is now allowed even if some underlying MyISAM tables are read-only. (Bug #702)
• Fixed a problem with CREATE TABLE t1 SELECT x'41'. (Bug #801)
• Removed some incorrect lock warnings from the error log.
• Fixed memory overrun when doing REPAIR on a table with a multi-part auto_increment key where one part was a packed CHAR.
• Fixed a probable race condition in the replication code that could potentially lead to INSERT statements not being replicated in the event of a FLUSH LOGS command or when the binary log exceeds max_binlog_size. (Bug #791)
• Fixed a crashing bug in INTERVAL and GROUP BY or DISTINCT. (Bug #807)
• Fixed bug in mysqlhotcopy so it actually aborts for unsuccessful table copying operations. Fixed another bug so that it succeeds when there are thousands of tables to copy. (Bug #812)
• Fixed problem with mysqlhotcopy failing to read options from option files. (Bug #808)
• Fixed bugs in optimiser that sometimes prevented MySQL from using FULLTEXT indexes even though it was possible (for example, in SELECT * FROM t1 WHERE MATCH a,b AGAINST("index") > 0).
• Fixed a bug with ``table is full'' in UNION operations.
• Fixed a security problem that enabled users with no privileges to obtain information on the list of existing databases by using SHOW TABLES and similar commands.
• Fixed a stack problem on UnixWare/OpenUnix.
• Fixed a configuration problem on UnixWare/OpenUNIX and OpenServer.
• Fixed a stack overflow problem in password verification.
• Fixed a problem with max_user_connections.
• HANDLER without an index now works properly when a table has deleted rows. (Bug #787)
• Fixed a bug with LOAD DATA in mysqlbinlog. (Bug #670).
• Fixed that SET CHARACTER SET DEFAULT works. (Bug #462)
• Fixed MERGE table behaviour in ORDER BY ... DESC queries. (Bug #515)
• Fixed server crash on PURGE MASTER LOGS or SHOW MASTER LOGS when the binary log is off. (Bug #733)
• Fixed password-checking problem on Windows. (Bug #464)
• Fixed the bug in comparison of a DATETIME column and an integer constant. (Bug #504)
• Fixed remote mode of mysqlbinlog. (Bug #672)
• Fixed ERROR 1105: Unknown error that occurred for some SELECT queries, where a column that was declared as NOT NULL was compared with an expression that took NULL value.
• Changed timeout in mysql_real_connect() to use poll() instead of select() to work around problem with many open files in the client.
• Fixed incorrect results from MATCH ... AGAINST used with a LEFT JOIN query.
• Fixed a bug that limited the maximum value for mysqld variables to 4294967295 when they are specified on the command line.
• Fixed a bug that sometimes caused spurious ``Access denied'' errors in HANDLER ... READ statements, when a table is referenced via an alias.
• Fixed portability problem with safe_malloc, which caused MySQL to give "Freeing wrong aligned pointer" errors on SCO 3.2.
• ALTER TABLE ... ENABLE/DISABLE KEYS could cause a core dump when done after an INSERT DELAYED statement on the same table.
• Fixed problem with conversion of localtime to GMT where some times resulted in different (but correct) timestamps. Now MySQL should use the smallest possible timestamp value in this case. (Bug #316)
• Very small query cache sizes could crash mysqld. (Bug #549)
• Fixed a bug (accidentally introduced by us but present only in version 4.0.13) that made INSERT ... SELECT into an AUTO_INCREMENT column not replicate well. This bug is in the master, not in the slave. (Bug #490)
• Fixed a bug: When an INSERT ... SELECT statement inserted rows into a non-transactional table, but failed at some point (for example, due to a ``Duplicate key'' error), the query was not written to the binlog. Now it is written to the binlog, with its error code, as all other queries are. About the slave-skip-errors option for how to handle partially completed queries in the slave, see section 4.10.6 Replication Options in `my.cnf'. (Bug #491)
• SET FOREIGN_KEY_CHECKS=0 was not replicated properly. The fix probably will not be backported to 3.23.
• On a slave, LOAD DATA INFILE which had no IGNORE or REPLACE clause on the master, was replicated with IGNORE. While this is not a problem if the master and slave data are identical (a LOAD that produces no duplicate conflicts on the master will produce none on the slave anyway), which is true in normal operation, it is better for debugging not to silently add the IGNORE. That way, you can get an error message on the slave and discover that for some reason, the data on master and slave are different and investigate why. (Bug #571)
• On a slave, LOAD DATA INFILE printed an incomplete ``Duplicate entry '%-.64s' for key %d''' message (the key name and value were not mentioned) in case of duplicate conflict (which does not happen in normal operation). (Bug #573)
• When using a slave compiled with --debug, CHANGE MASTER TO RELAY_LOG_POS could cause a debug assertion failure. (Bug #576)
• When doing a LOCK TABLES WRITE on an InnoDB table, commit could not happen, if the query was not written to the binary log (for example, if --log-bin was not used, or binlog-ignore-db was used). (Bug #578)
• If a 3.23 master had open temporary tables that had been replicated to a 4.0 slave, and the binlog got rotated, these temporary tables were immediately dropped by the slave (which caused problems if the master used them subsequently). This bug had been fixed in 4.0.13, but in a manner which caused an unlikely inconvenience: if the 3.23 master died brutally (power failure), without having enough time to automatically write DROP TABLE statements to its binlog, then the 4.0.13 slave would not notice the temporary tables have to be dropped, until the slave mysqld server is restarted. This minor inconvenience is fixed in 3.23.57 and 4.0.14 (meaning the master must be upgraded to 3.23.57 and the slave to 4.0.14 to remove the inconvenience). (Bug #254)
• If MASTER_POS_WAIT() was waiting, and the slave was idle, and the SQL slave thread terminated, MASTER_POS_WAIT() would wait forever. Now when the SQL slave thread terminates, MASTER_POS_WAIT() immediately returns NULL (``slave stopped''). (Bug #651)
• After RESET SLAVE; START SLAVE;, the Relay_log_space value displayed by SHOW SLAVE STATUS was too big by four bytes. (Bug #763)
• If a query was ignored on the slave (because of replicate-ignore-table and other similar rules), the slave still checked if the query got the same error code (0, no error) as on the master. So if the master had an error on the query (for example, ``Duplicate entry'' in a multiple-row insert), then the slave stopped and warned that the error codes didn't match. (Bug #797)

User Comments

# Fixed a bug (accidentally introduced by us, only in version 4.0.13) which made INSERT ... SELECT in an auto_increment column not replicate well. This bug is in the master, not in the slave. Bug #490.

This changelog line is kind of misleading. If you have a master->slave set up, running an INSERT ... SELECT will not, in my experience replicate. It will generate a duplicate key error on the slave, because the master will binlog a SET INSERT_ID=##; statement where it doesn't belong, which will lead to the slave stopping when you hit a duplicate key.

Could you add a 'PURGE RELAY LOGS' command similar to 'PURGE MASTER LOGS' ?

Or is there already a way of doing that ?

Add your own comment.