Functionality added or changed:
Incompatible Change: Due to a change in the naming scheme for partitioning and subpartitioning files, it is not possible for the server to read partitioned tables created in previous MySQL versions. Attempting to read pre-5.1.6 partitioned tables with a MySQL 5.1.7 or later server now generates a suitable warning message.
Two possible workarounds are:
Create a nonpartitioned table with the same table
schema using a standard
TABLE statement (that is, with no
SELECT INTO to copy the
data into the nonpartitioned table before the
Following the upgrade, you can partition the new table
ALTER TABLE ... PARTITION BY ....
In either case, you should drop the pre-5.1.6 partitioned tables before upgrading to 5.1.6 or later.
If any partitioned tables that were created prior to MySQL
5.1.6 are present following an upgrade to MySQL 5.1.6 or
later, it is also not possible to read from the
table, nor will you be able to drop those tables or the
database or databases in which they are located. In this
event, you must:
Shut down mysqld
Manually delete the table, partition, and (if any) subpartition files
Restart the MySQL Server
TYPE = is no longer accepted as a synonym for the
ENGINE = table option. (
TYPE has been
deprecated since MySQL 4.0.)
SELECT ... FROM
INFORMATION_SCHEMA.FILES now raises a warning in the
event that the cluster has crashed.
In row-based replication, when executing a Rows_log_event, the
associated table was locked, the rows applied and the lock
released. This did not work since there are storage engines that
count locks and perform an autocommit when the number of locks
reach zero. Now we ensure that all table maps come before all
ROWS events in a statement.
Disk Data: Status messages have been added to ndb_restore to enable users to know that data files for Disk Data are being created. (Bug#16873)
It is now possible to replicate
tables having no explicit primary key. See
Section 17.6, “MySQL Cluster Replication”.
Creator privileges are now checked for all events before execution. (Bug#17289)
The SQL mode in effect at the time an event is created or altered is recorded and used during event execution. (Bug#16407)
--wait-timeout option to
mysqlmanager to allow configuration of the
timeout for dropping an inactive connection, and increased the
default timeout from 30 seconds to 28,800 seconds (8 hours).
All subpartitions within a given partitioned table are now guaranteed to have unique names. (Bug#15408)
RENAME DATABASE statement.
PROCESSLIST table to
Several changes were made to make upgrades easier:
Added the mysql_upgrade program that checks all tables for incompatibilities with the current version of MySQL Server and repairs them if necessary. This program should be run for each MySQL upgrade (rather than mysql_fix_privilege_tables). See Section 4.4.8, “mysql_upgrade — Check Tables for MySQL Upgrade”.
FOR UPGRADE option for the
CHECK TABLE statement. This
option checks whether tables are incompatible with the
current version of MySQL Server.
IN NATURAL LANGUAGE MODE and
IN NATURAL LANGUAGE MODE WITH QUERY EXPANSION
modifiers for full-text searches. See
Section 11.8, “Full-Text Search Functions”.
MySQL Cluster: ndbd restarts could sometimes fail due to incorrect memory access. (Bug#17417)
MySQL Cluster: Sharing of table names containing special characters between multiple SQL nodes was not handled correctly when binary logging was enabled (a timeout error resulted). (Bug#17415)
MySQL Cluster: Table definitions were not shared between multiple SQL nodes in a cluster without binary logging being enabled. (Bug#17414)
Cluster log file paths were truncated to 128 characters. They
may now be as long as
MAX_PATH (the maximum
path length permitted by the operating system).
REDO log would become corrupted (and thus
unreadable) in some circumstances, due to a failure in the query
MySQL Cluster: The storage engine did not allow views to be updated. (Bug#17206)
CREATE TABLE failed when
old_tbl used the
NDB storage engine.
An unhandled resources issue could cause node failure with a
DELETE FROM TABLE affecting thousands of
UNIQUE keys in Cluster tables were limited to
225 bytes in length.
No error message was generated for setting
NoOfFragmentLogFiles too low.
No error message was generated for setting
MaxNoOfAttributes too low.
MySQL Cluster: Performing large numbers of data manipulation statements on cluster tables using Disk Data could lead to a server crash.
For a transaction that used
InnoDB tables, interruption of the
transaction due to a dropped connection on a master server
caused slaves to lose synchrony.
Previously, a stored function invocation was written to the
binary log as
invocation changes data and occurs within a nonlogged statement,
or if the function invokes a stored procedure that produces an
error. These invocations now are logged as
better control over error code checking (slave servers could
stop due to detecting a different error than occurred on the
Disk Data: In some cases, a cluster using Disk Data tables could not be restarted following a normal shutdown. (Bug#16872)
Row-based replication was not set up correctly if a backup was
already in progress. For example, connecting a
mysqld instance to a cluster which was being
backed up would result in the message NDB: skipping
being written to the error log.
Cluster Replication: Cluster tables not having an explicit primary key could not be replicated. (Bug#14541)
Column counts were encoded incorrectly in the binary log for row-based logging format. (Bug#17678)
Data truncations on non-
UNIQUE indexes could
InnoDB when using multi-byte character
Execution times for scheduled events were not calculated correctly: the last execution time was used as a base rather than the actual start time. (Bug#17494)
Race conditions between event creation, dropping, and execution could result in a server crash or hang. (Bug#17373)
Trying to create a partitioned table with more than 32 attributes failed. (Bug#17179)
Attempting to add a new partition to a table partitioned by a unique key would cause an Out of memory error. (Bug#17169)
On Windows platforms, some attempts to create partitioned tables
from the command line would cause the
client to hang.
Statements that contained Unicode characters were not logged to the log tables correctly. (Bug#16905)
REPLACE INTO on a partitioned table
having a primary key would crash the server in the event of a
duplicate key error.
If the query optimizer transformed a
clause in a subquery, it did not also transform the
HAVING clause if there was one, producing
on a nonmax server caused a server crash. This also happened
following the creation of a table with a very large number
(hundreds) of partitions.
GROUP BY on column used in
WHERE clause could cause empty set to be
A statement containing
GROUP BY and
HAVING clauses could return incorrect results
HAVING clause contained logic that
FALSE for every row.
Killing a long-running query containing a subquery could cause a server crash. (Bug#14851)
SET sql_mode = ,
N > 31, did not work
InnoDB could display an incorrect error
message for a cascading update.
Repeated invocation of
my_end() caused corruption of character set
data and connection failure.