MySQL 5.6 Release Notes
This is a milestone release, for use at your own risk. Upgrades between milestone releases (or from a milestone release to a GA release) are not supported. Significant development changes take place in milestone releases and you may encounter compatibility issues, such as data format changes that require attention in addition to the usual procedure of running mysql_upgrade. For example, you may find it necessary to dump your data with mysqldump before the upgrade and reload it afterward. (Making a backup before the upgrade is a prudent precaution in any case.)
Beginning with MySQL 5.6.5, Oracle no longer provides binaries for OS X 10.5. This aligns with Apple no longer providing updates or support for this platform.
Previously, at most one TIMESTAMP
column per table could be automatically initialized or updated
to the current date and time. This restriction has been lifted.
Any TIMESTAMP
column definition
can have any combination of DEFAULT
CURRENT_TIMESTAMP
and ON UPDATE
CURRENT_TIMESTAMP
clauses. In addition, these clauses
now can be used with DATETIME
column definitions. For more information, see
Automatic Initialization and Updating for TIMESTAMP and DATETIME.
Important Change; Replication:
This release introduces global
transaction identifiers (GTIDs) for MySQL
Replication. A GTID is a unique identifier that is assigned to
each transaction as it is committed; this identifier is unique
on the MySQL Server where the transaction originated, as well as
across all MySQL Servers in a given replication setup. Because
GTID-based replication depends on tracking transactions, it
cannot be employed with tables that employ a nontransactional
storage engine such as MyISAM
;
thus, it is currently supported only with
InnoDB
tables.
Because each transaction is uniquely identified, it is not
necessary when using GTIDs to specify positions in the master's
binary log when starting a new slave or failing over to a new
master. This is reflected in the addition of a new
MASTER_AUTO_POSITION
option for the
CHANGE MASTER TO
statement which
takes the place of the MASTER_LOG_FILE
and
MASTER_LOG_POS
options when executing this
statement to prepare a MySQL Server to act as a replication
slave.
To enable GTIDs on a MySQL Server, the server must be started
with the options --gtid-mode=ON
--disable-gtid-unsafe-statements
--log-bin
--log-slave-updates
. These
options are needed whether the server acts as a replication
master or as a replication slave; the
--gtid-mode
and
--disable-gtid-unsafe-statements
options are
new in this release. Once the master and slave have each been
started with these options, it is necessary only to issue a
CHANGE MASTER TO ... MASTER_AUTO_POSITION=1
followed by START SLAVE
on the
slave to start replication.
A number of new server system variables have also been added for monitoring GTID usage. For more information about these options and variables, see Global Transaction ID Options and Variables.
As part of these changes, three new
mysqlbinlog
options—--include-gtids
,
--exclude-gtids
, and
--skip-gtids
—have been
added for reading binary logs produced when the server
participates in replication with GTIDs.
Due to an issue discovered just prior to release, you cannot
import a dump made using mysqldump from a
MySQL 5.5 server to a MySQL 5.6.5 server and then use
mysqlupgrade on the MySQL 5.6.5 server
while GTIDs are enabled; doing so makes it impossible to
connect to the server normally following the upgrade. Instead,
you should import the dump and run
mysqlupgrade while the MySQL 5.6.5 server
is running with
--gtid-mode=OFF
, then restart
it with --gtid-mode=ON
. (Bug #13833710)
(mysqlupgrade can be executed when the
server is running with
--gtid-mode
set either to
OFF
, or to ON
.)
For additional information about GTIDs and setting up GTID-based replication, see Replication with Global Transaction Identifiers.
MySQL now provides improved access to the host cache, which contains client IP address and host name information and is used to avoid DNS lookups, as well as more information about the causes of errors that occur when clients connect to the server. These changes have been implemented:
A new Performance Schema
host_cache
table exposes the
contents of the host cache so that it can be examined using
SELECT
statements. Access to
host cache contents makes it possible to answer questions
such as how many hosts are cached, what kinds of connection
errors are occurring for which hosts, or how close host
error counts are to reaching the
max_connect_errors
system
variable limit. The Performance Schema must be enabled or
this table is empty.
If you upgrade to this MySQL release from an earlier
version, you must run mysql_upgrade (and
restart the server) to incorporate this change into the
performance_schema
database.
The host cache has additional counters to track errors that do apply to specific IP addresses.
The host cache size now is configurable using the
host_cache_size
system
variable. This variable also controls the maximum size of
the host_cache
table because
that table is the cache visible representation. Setting the
size to 0 disables the host cache. This is similar to
disabling the cache by starting the server with
--skip-host-cache
, but using
host_cache_size
is more
flexible because it can also be used to resize, enable, and
disable the host cache at runtime, not just at server
startup. If you start the server with
--skip-host-cache
, the host
cache cannot be re-enabled at runtime.
New
Connection_errors_
status variables provide information about connection errors
that do not apply to specific client IP addresses.
xxx
For more information, see DNS Lookups and the Host Cache, and The host_cache Table. (Bug #22821, Bug #24906, Bug #45817, Bug #59404, Bug #11746048, Bug #11746269, Bug #11754244, Bug #11766316)
Reporting of how to sort a result set in
EXPLAIN
has been improved for
some statements. This sorting decision could be reported
incorrectly, causing Using filesort
or
Using temporary
to be reported when they
should not have been or vice versa. This could occur for
statements that included index hints, that had the form
SELECT SQL_BIG_RESULT ... GROUP BY
, that used
SQL_CALC_FOUND_ROWS
with
LIMIT
, or that used GROUP
BY
, ORDER BY
, and
LIMIT
.
(Bug #11744768, Bug #1560)
These query optimizer improvements were implemented:
The EXPLAIN
statement now can
produce output in JSON format. To select this, use
EXPLAIN FORMAT =
JSON
syntax. With explainable_stmt
FORMAT = JSON
, the output
includes regular EXPLAIN
information, as well as extended and partition information.
Traditional EXPLAIN
output
has also changed so that empty columns contain
NULL
rather the empty string. In
addition, UNION RESULT
rows have
Using filesort
in the
Extra
column because a temporary table is
used to buffer UNION
results.
To work for both Optimizer Trace and JSON-format
EXPLAIN
output, the
end_marker
parameter for the
optimizer_trace
system
variable has been moved to a separate
end_markers_in_json
system
variable. This is an incompatible change to the
optimizer_trace
variable.
For more information, see
MySQL
Internals: Tracing the Optimizer.
The optimizer tries to find the best query execution plan by beginning with the most promising table and recursively adding to the plan the most promising of the remaining tables. Partial execution plans with a higher cost than an already found plan are pruned. The optimizer now attempts to improve the order in which it adds tables to the plan, resulting in a reduction of the number of partial plans considered.
Queries that are likely to have improved performance are
joins of many tables, where most tables use
eq_ref
or
ref
join types (as
indicated by EXPLAIN
output).
A new status variable,
Last_query_partial_plans
,
counts the number of iterations the optimizer makes in
execution plan construction for the previous query.
The optimizer uses semijoin and materialization strategies to optimize subquery execution. See Optimizing Subqueries with Semijoin Transformations, and Optimizing Subqueries with Materialization. In addition, the Batched Key Access (BKA) Join and Block Nested-Loop (BNL) Join algorithms used for inner join and outer join operations have been extended to support semijoin operations. For more information, see Block Nested-Loop and Batched Key Access Joins.
Several flags have been added to the
optimizer_switch
system
variable to enable control over semijoin and subquery
materialization strategies. The
semijoin
flag controls
whether semijoins are used. If it is set to
on
, the
firstmatch
and
loosescan
flags enable
finer control over the permitted semijoin strategies. The
materialization
flag
controls whether subquery materialization is used. If
semijoin
and
materialization
are both
on
, semijoins also use materialization
where applicable. These flags are on
by
default. See Switchable Optimizations.
For expressions such as
that compare
a column to a list of values, the optimizer previously made
row estimates using index dives for each value in the list.
This becomes inefficient as the number of values becomes
large. The optimizer now can make row estimates for such
expressions using index statistics instead, which is less
accurate but quicker for a large number of values. The point
at which the optimizer switches from index dives to index
statistics is configurable using the new
col_name
IN(values
)eq_range_index_dive_limit
system variable. For more information, see
Equality Range Optimization of Many-Valued Comparisons.
The Performance Schema has these additions:
The Performance Schema now has a
host_cache
table that exposes
the contents of the host cache so that it can be examined
using SELECT
statements. See
Host Cache Notes elsewhere in this changelog.
The Performance Schema now maintains statement digest information. This normalizes and groups statements with the same “signature” and permits questions to be answered about the types of statements the server is executing and how often they occur.
A statement_digest
consumer in the
setup_consumers
table
controls whether the Performance Schema maintains digest
information.
The statement event tables
(events_statements_current
,
events_statements_history
,
and
events_statements_history_long
)
have DIGEST
and
DIGEST_TEXT
columns that contain
digest MD5 values and the corresponding normalized
statement text strings.
A
events_statements_summary_by_digest
table provides aggregated statement digest information.
For more information, see The host_cache Table, Performance Schema Statement Event Tables, and Statement Summary Tables.
If you upgrade to this MySQL release from an earlier version,
you must run mysql_upgrade (and restart the
server) to incorporate these changes into the
performance_schema
database.
Passwords stored in the older hash format used before MySQL 4.1
are less secure than passwords that use the native password
hashing method and should be avoided. Pre-4.1 passwords and the
mysql_old_password
authentication plugin are
now deprecated. To prevent connections using accounts that have
pre-4.1 password hashes, the
secure_auth
system variable is
now enabled by default. (To permit connections for accounts that
have such password hashes, start the server with
--secure_auth=0
.)
(Bug #13586336)
MySQL client programs now issue a warning if a password is given on the command line that this can be insecure.
Incompatible Change:
The obsolete OPTION
modifier for the
SET
statement has been removed.
InnoDB:
--ignore-builtin-innodb
is now
ignored if used.
(Bug #13586262)
OS X; Microsoft Windows:
A new CMake option,
MYSQL_PROJECT_NAME
, can be set on
Windows or OS X to be used in the project name.
(Bug #13551687)
Microsoft Windows:
A new server option,
--slow-start-timeout
, controls
the Windows service control manager's service start timeout. The
value is the maximum number of milliseconds that the service
control manager waits before trying to kill the MySQL service
during startup. The default value is 15000 (15 seconds). If the
MySQL service takes too long to start, you may need to increase
this value. A value of 0 means there is no timeout.
(Bug #45546, Bug #11754011)
The MySQL-shared-compat
RPM package enables
users of Red Hat-provided mysql-*-5.1
RPM
packages to migrate to Oracle-provided
MySQL-*-5.5
packages.
MySQL-shared-compat
now replaces the Red Hat
mysql-libs
package by replacing
libmysqlclient.so
files of the latter
package, thus satisfying dependencies of other packages on
mysql-libs
. This change affects only users of
Red Hat (or Red Hat-compatible) RPM packages. Nothing is
different for users of Oracle RPM packages.
(Bug #13867506)
Temporary tables for INFORMATION_SCHEMA
queries now use dynamic MyISAM
row
format if they contain sufficiently large
VARCHAR
columns, resulting in
space savings.
(Bug #13627632)
As of MySQL 5.5.3, the LOW_PRIORITY
modifier
for LOCK TABLES ...
LOW_PRIORITY WRITE
has no effect. This modifier is now
deprecated. Its use should be avoided and now produces a
warning. Use LOCK
TABLES ... WRITE
instead.
(Bug #13586314)
If the
log_queries_not_using_indexes
system variable is enabled, slow queries that do not use indexes
are written to the slow query log. In this case, it is now
possible to put a logging rate limit on these queries by setting
the new
log_throttle_queries_not_using_indexes
system variable, so that the slow query log does not grow too
quickly. By default, this variable is 0, which means there is no
limit. Positive values impose a per-minute limit on logging of
queries that do not use indexes. The first such query opens a
60-second window within which the server logs queries up to the
given limit, then suppresses additional queries. If there are
suppressed queries when the window ends, the server logs a
summary that indicates how many there were and the aggregate
time spent in them. The next 60-second window begins when the
server logs the next query that does not use indexes.
(Bug #55323, Bug #11762697)
Several subquery performance issues were resolved through the implementation of semijoin subquery optimization strategies. See Optimizing Subqueries with Semijoin Transformations. (Bug #47914, Bug #11756048, Bug #58660, Bug #11765671, Bug #10815, Bug #11745162, Bug #9021, Bug #13519134, Bug #48763, Bug #11756798, Bug #25130, Bug #11746289)
The mysql client now supports an
--init-command=
option. The option value is an SQL statement to execute after
connecting to the server. If auto-reconnect is enabled, the
statement is executed again after reconnection occurs.
(Bug #45634, Bug #11754087)str
New utf8_general_mysql500_ci
and
ucs2_general_mysql500_ci
collations have been
added that preserve the behavior of
utf8_general_ci
and
ucs2_general_ci
from versions of MySQL
previous to 5.1.24. Bug #27877 corrected an error in the
original collations but introduced an incompatibility for
columns that contain German 'ß'
LATIN SMALL
LETTER SHARP S. (As a result of the fix, that character compares
equal to characters with which it previously compared
different.) A symptom of the problem after upgrading to MySQL
5.1.24 or newer from a version older than 5.1.24 is that
CHECK TABLE
produces this error:
Table upgrade required. Please do "REPAIR TABLE `t`" or dump/reload to fix it!
Unfortunately, REPAIR TABLE
could
not fix the problem. The new collations permit older tables
created before MySQL 5.1.24 to be upgraded to current versions
of MySQL.
To convert an affected table after a binary upgrade that leaves
the table files in place, alter the table to use the new
collation. Suppose that the table t1
contains
one or more problematic utf8
columns. To
convert the table at the table level, use a statement like this:
ALTER TABLE t1 CONVERT TO CHARACTER SET utf8 COLLATE utf8_general_mysql500_ci;
To apply the change on a column-specific basis, use a statement
like this (be sure to repeat the column definition as originally
specified except for the COLLATE
clause):
ALTER TABLE t1 MODIFY c1 CHAR(N) CHARACTER SET utf8 COLLATE utf8_general_mysql500_ci;
To upgrade the table using a dump and reload procedure, dump the
table using mysqldump, modify the
CREATE TABLE
statement in the
dump file to use the new collation, and reload the table.
After making the appropriate changes, CHECK
TABLE
should report no error.
(Bug #43593, Bug #11752408)
References: See also: Bug #27877.
The SET TRANSACTION
and
START
TRANSACTION
statements now support READ
WRITE
and READ ONLY
modifiers to
set the transaction access mode for tables used in transactions.
The default mode is read/write, which is the same mode as
previously. Read/write mode now may be specified explicitly with
the READ WRITE
modifier. Using READ
ONLY
prohibits table changes and may enable storage
engines to make performance improvements that are possible when
changes are not permitted.
In addition, the new
--transaction-read-only
option
and tx_read_only
system
variable permit the default transaction access mode to be set at
server startup and runtime.
For more information, see SET TRANSACTION Statement, and START TRANSACTION, COMMIT, and ROLLBACK Statements.
MySQL distributions no longer include the GPL
readline
input-editing library. This results
in simpler maintenance and support, and simplifies licensing
considerations.
Incompatible Change; Replication:
CHANGE MASTER TO
statements were
written into the error log using quoted numeric values, although
the syntax for this statement does not allow such option values
to be quoted. This meant that such statements could not be
copied from the error log and re-run verbatim. Now
CHANGE MASTER TO
statements are
written to the error log without the extraneous quotation marks,
and so are syntactically correct as logged.
Incompatible Change:
A change in MySQL 5.6.3 caused
LAST_DAY()
to be more strict and
reject incomplete dates with a day part of zero. For this
function, a nonzero day part is not necessary, so the change has
been reverted.
(Bug #13458237)
Important Change; InnoDB:
When a row grew in size due to an UPDATE
operation, other (non-updated) columns could be moved to
off-page storage so that information about the row still fit
within the constraints of the InnoDB
page
size. The pointer to the new allocated off-page data was not set
up until the pages were allocated and written, potentially
leading to lost data if the system crashed while the column was
being moved out of the page. The problem was more common with
tables using ROW_FORMAT=DYNAMIC
or
ROW_FORMAT=COMPRESSED
along with the
Barracuda file format, particularly with the
innodb_file_per_table
setting
enabled, because page allocation operations are more common as
the .ibd
tablespace files are extended.
Still, the problem could occur with any combination of InnoDB
version, file format, and row format.
A related issue was that during such an
UPDATE
operation, or an
INSERT
operation that reused a delete-marked
record, other transactions could see invalid data for the
affected column, regardless of isolation level.
The fix corrects the order of operations for moving the column data off the original page and replacing it with a pointer. Now if a crash occurs at the precise moment when the column data is being transferred, the transfer will not be re-run during crash recovery.
In MySQL 5.1, this fix applies to the InnoDB Plugin, but not the built-in InnoDB storage engine. (Bug #13721257, Bug #12612184, Bug #12704861)
Important Change; Partitioning: The query cache did not always function correctly with partitioned tables in a transactional context. For this reason, the query cache is now disabled for any queries using partitioned tables, and such queries can no longer be cached. For more information, see Restrictions and Limitations on Partitioning. (Bug #11761296, Bug #53775)
Important Change; Replication:
The CHANGE MASTER TO
statement
was not checked for invalid characters in values for options
such as MASTER_HOST
and
MASTER_USER
. In addition, when the server was
restarted, a value containing certain characters was trimmed,
causing the loss of its original value. Now such values are
validated, and in cases where the value contains invalid
characters, including linefeed (\n
or
0x0A
) characters, the statement fails with an
error (ER_MASTER_INFO).
(Bug #11758581, Bug #50801)
Important Change; Replication:
Moving the binary log file, relay log file, or both files to a
new location, then restarting the server with a new value for
--log-bin
,
--relay-log
, or both, caused the
server to abort on start. This was because the entries in the
index file overrode the new location. In addition, paths were
calculated relative to datadir (rather than to the
--log-bin
or
--relay-log
values).
The fix for this problem means that, when the server reads an
entry from the index file, it now checks whether the entry
contains a relative path. If it does, the relative part of the
path is replaced with the absolute path set using the
--log-bin
or
--relay-log
option. An absolute
path remains unchanged; in such a case, the index must be edited
manually to enable the new path or paths to be used.
(Bug #11745230, Bug #12133)
Performance; InnoDB:
The optimizer now takes into account InnoDB
page sizes other than 16KB, which can be configured with the
innodb_page_size
option when
creating a MySQL instance. This change improves the estimates of
I/O costs for queries on systems with non-default
InnoDB
page sizes.
(Bug #13623078)
Performance; InnoDB:
Memory allocation for InnoDB
tables was
reorganized to reduce the memory overhead for large numbers of
tables or partitions, avoiding situations where the
“resident set size” could grow regardless of
FLUSH TABLES
statements. The problem was most
evident for tables with large row size. Some of the memory that
was formerly allocated for every open table is now allocated
only when the table is modified for the first time.
(Bug #11764622, Bug #57480)
Performance:
Temporary MyISAM
tables (unlike
normal MyISAM
tables) did not use the dynamic
row format when they contained
VARCHAR
columns, resulting in
larger temporary files (and more file I/O) than necessary.
Dynamic row format now is used, which results in smaller tables
that are faster to process.
(Bug #13350136, Bug #78840, Bug #22023218)
InnoDB:
An erroneous assertion could occur, in debug builds only, when
creating an index on a column containing zero-length values
(that is, ''
).
(Bug #13654923)
InnoDB:
A DDL operation such as ALTER TABLE ... ADD
COLUMN
could stall, eventually timing out with an
Error 1005: Can't create table
message
referring to fil_rename_tablespace
.
(Bug #13636122, Bug #62100, Bug #63553)
InnoDB:
If InnoDB was started with
innodb_force_recovery
set to a value of 3 or
4, and there are
transactions to
roll back, normal
shutdown would hang waiting
for those transactions to complete. Now the shutdown happens
immediately, without rolling back any transactions, because
nonzero values for innodb_force_recovery
are
only appropriate for troubleshooting and diagnostic purposes.
(Bug #13628420)
InnoDB:
The MySQL server could hang in some cases if the configuration
option innodb_use_native_aio
was turned off.
(Bug #13619598)
InnoDB:
A Valgrind error was fixed in the function
os_aio_init()
.
(Bug #13612811)
InnoDB:
The configuration option innodb_sort_buf_size
was renamed to
innodb_sort_buffer_size
for
consistency. This work area is used while creating an
InnoDB
index.
(Bug #13610358)
InnoDB:
The server could crash when creating an
InnoDB
temporary table under Linux, if the
$TMPDIR
setting points to a
tmpfs
filesystem and
innodb_use_native_aio
is
enabled, as it is by default in MySQL 5.5.4 and higher. The
entry in the error log looked like:
101123 2:10:59 InnoDB: Operating system error number 22 in a file operation. InnoDB: Error number 22 means 'Invalid argument'.
The crash occurred because asynchronous I/O is not supported on
tmpfs in some Linux kernel versions. The workaround was to turn
off the innodb_use_native_aio
setting or use
a different temporary directory. The fix causes
InnoDB
to turn off the
innodb_use_native_aio
setting automatically
if it detects that the temporary file directory does not support
asynchronous I/O.
(Bug #13593888, Bug #11765450, Bug #58421)
InnoDB:
During startup, the status variable
Innodb_buffer_pool_dump_status
could be empty for a brief time before being initialized to the
correct value not started
.
(Bug #13513676)
InnoDB:
Valgrind errors when referencing the internal function
buf_LRU_scan_and_free_block()
were fixed.
(Bug #13491704)
InnoDB: The MySQL error log could contain messages like:
InnoDB: Ignoring strange row from mysql.innodb_index_stats WHERE ...
The fix makes the contents of the
innodb_index_stats
and
innodb_table_stats
tables case-sensitive, to
properly distinguish the statistics for tables whose names
differ only in lettercase. Other cases were fixed where the
wrong name could be selected for an index while retrieving
persistent statistics.
(Bug #13432465)
InnoDB:
References to C preprocessor symbols and macros
HAVE_purify
,
UNIV_INIT_MEM_TO_ZERO
, and
UNIV_SET_MEM_TO_ZERO
were removed from the
InnoDB
source code. They were only used in
debug builds instrumented for Valgrind. They are replaced by
calls to the UNIV_MEM_INVALID()
macro.
(Bug #13418934)
InnoDB: The MySQL server could halt with an assertion error:
InnoDB: Failing assertion: page_get_n_recs(page) > 1
Subsequent restarts could fail with the same error. The error
occurred during a purge
operation involving the InnoDB
change buffer. The
workaround was to set the configuration option
innodb_change_buffering=inserts
.
(Bug #13413535, Bug #61104)
InnoDB:
A discrepancy could arise between the number of available
InnoDB
undo
logs and the number of undo logs that were currently
active. Now the
innodb_undo_logs
system
variable reports the number of active undo logs, and the new
Innodb_available_undo_logs
status variable reports the total number of undo logs.
(Bug #13255225)
InnoDB:
When doing a live downgrade from MySQL 5.6.4 or later, with
innodb_page_size
set to a value other than
16384, now the earlier MySQL version reports that the page size
is incompatible with the older version, rather than crashing or
displaying a “corruption” error.
(Bug #13116225)
InnoDB:
Certain CREATE TABLE
statements
could fail for InnoDB
child tables containing
foreign key definitions. This problem affected Windows systems
only, with the setting
lower_case_table_names=0
. It was a regression
from MySQL bug #55222.
(Bug #13083023, Bug #60229)
InnoDB:
If the server crashed during a TRUNCATE
TABLE
or CREATE INDEX
statement for an InnoDB
table, or a
DROP DATABASE
statement for a
database containing InnoDB
tables, an index
could be corrupted, causing an error message when accessing the
table after restart:
InnoDB: Error: trying to load indexindex_name
for tabletable_name
InnoDB: but the index tree has been freed!
In MySQL 5.1, this fix applies to the InnoDB Plugin, but not the built-in InnoDB storage engine. (Bug #12861864, Bug #11766019)
InnoDB:
A DDL operation for an InnoDB
table could
cause a busy MySQL server to halt with an assertion error:
InnoDB: Failing assertion: trx->error_state == DB_SUCCESS
The error occurred if the DDL operation was run while all 1023
undo slots were in use by concurrent transactions. This error
was less likely to occur in MySQL 5.5 and 5.6, because raising
the number of InnoDB
undo slots increased the
number of simultaneous transactions (corresponding to the number
of undo slots) from 1K to 128K.
(Bug #12739098, Bug #62401)
InnoDB:
InnoDB
persistent statistics gave less
accurate estimates for date columns than for columns of other
data types. The fix changes the way cardinality is estimated for
nonunique keys, and avoids situations where identical values
could be counted twice if they occurred on different index
pages.
(Bug #12429443)
InnoDB:
The innodb_max_purge_lag
variable controls how to delay DML operations when purge
operations are lagging. Previously, if an old consistent read
view was detected, DML operations would not be delayed even
though the purge lag exceeded the
innodb_max_purge_lag
setting.
Additionally, if the
innodb_max_purge_lag
setting
was used, situations could arise in which the DML delay time
would continue to increase but not be applied right away due to
the presence an old consistent read view. This could result in a
lengthy DML delay when the accumulated DML delay time is
eventually applied.
This fix caps the DML delay at a maximum value, removes the consistent read check, and revises the DML delay calculation. (Bug #12407434, Bug #60776)
InnoDB:
With 1024 concurrent InnoDB
transactions
running concurrently and the
innodb_file_per_table
setting
enabled, a CREATE TABLE
operation
for an InnoDB
table could fail. The
.ibd
file from the failed CREATE
TABLE
was left behind, preventing the table from being
created later, after the load had dropped.
The fix adds error handling to delete the erroneous
.ibd
file. This error was less likely to
occur in MySQL 5.5 and 5.6, because raising the number of
InnoDB
undo slots increased the number of
simultaneous transactions needed to trigger the bug, from 1K to
128K.
(Bug #12400341)
InnoDB:
Improved the accuracy of persistent InnoDB
statistics for large tables. The estimate of distinct records
could be inaccurate if the index tree was more than 3 levels
deep.
(Bug #12316365)
InnoDB: Shutdown could hang with messages like this in the log:
Waiting for purge thread to be suspended
After 1 hour, the shutdown times out and
mysqld
quits. This problem is most likely to
occur with a high value for
innodb_purge_threads
.
(Bug #11765863, Bug #58868, Bug #60939)
InnoDB:
When DROP TABLE
failed due to all
undo slots being in use, the error returned was
Unknown table '...' rather than the
expected Too many active concurrent
transactions.
(Bug #11764724, Bug #57586)
References: See also: Bug #11764668, Bug #57529.
InnoDB:
Server startup could produce an error for temporary tables using
the InnoDB
storage engine, if the path in the
$TMPDIR
variable ended with a
/
character. The error log would look like:
120202 19:21:26 InnoDB: Operating system error number 2 in a file operation. InnoDB: The error means the system cannot find the path specified. InnoDB: If you are installing InnoDB, remember that you must create InnoDB: directories yourself, InnoDB does not create them. 120202 19:21:26 InnoDB: Error: trying to open a table, but could not InnoDB: open the tablespace file './t/#sql7750_1_0.ibd'! InnoDB: Have you moved InnoDB .ibd files around without using the InnoDB: commands DISCARD TABLESPACE and IMPORT TABLESPACE? InnoDB: It is also possible that this is a temporary table #sql..., InnoDB: and MySQL removed the .ibd file for this.
The workaround for the problem was to create a similar temporary
table again, copy its .frm
file to
tmpdir
under the name mentioned in the error
message (for example, #sql123.frm
) and
restart mysqld
with tmpdir
set to its normal value without a trailing slash, for example
/var/tmp
. On startup, MySQL would see the
.frm
file and issue DROP
TABLE
for the orphaned temporary table.
(Bug #11754376, Bug #45976)
Partitioning:
When creating a view from a
SELECT
statement that used
explicit partition selection, the partition selection portion of
the query was ignored.
(Bug #13559657)
Partitioning:
Adding a partition to an already existing
LIST
-partitioned table did not work correctly
if the number of items in the new partition was greater than 16.
This could happen when trying to add a partition using an
ALTER
TABLE ... ADD PARTITION
statement, or an
ALTER
TABLE ... REORGANIZE PARTITION
statement.
This 16-item limit was not apparent when using either
CREATE TABLE ...
PARTITION BY LIST
or
ALTER
TABLE ... PARTITION BY LIST
.
(Bug #13029508, Bug #62505)
Partitioning: A function internal to the code for finding matching subpartitions represented an unsigned number as signed, with the result that matching subpartitions were sometimes missed in results of queries. (Bug #12725206, Bug #61765)
References: See also: Bug #20257.
Partitioning:
An
ALTER
TABLE ... ADD PARTITION
statement subsequent to
ALTER
TABLE ... REORGANIZE PARTITION
failed on a table
partitioned by HASH
or
KEY
.
(Bug #11764110, Bug #56909)
Replication:
Executing mysqlbinlog with the
--start-position=
option, where N
N
was equal either to 0
or to a value greater than the length of the dump file, caused
it to crash.
This issue was introduced in MySQL 5.5.18 by the fix for Bug #32228 and Bug #11747416. (Bug #13593869, Bug #64035)
References: This issue is a regression of: Bug #32228, Bug #11747416.
Replication:
When starting the server, replication repositories were checked
even when the server_id
system
variable was equal to 0 (the default), in spite of the fact that
a valid nonzero value for
server_id
must be supplied for
a server that acts as either a master or a slave in MySQL
replication.
This could cause problems when trying to perform a live upgrade
from MySQL 5.5, although it was possible to work around the
issue by starting the server with
--skip-slave-start
(in addition
to any other required options).
To avoid this problem, replication repositories are now checked
only when the server is started with
--server-id
using a nonzero value.
(Bug #13427444, Bug #13504821)
Replication:
Formerly, the default value shown for the
Port
column in the output of
SHOW SLAVE HOSTS
was 3306 whether
the port had been set incorrectly or not set at all. Now, when
the slave port is not set, the actual port used by the slave is
shown. This change also affects the default shown for the
--report-port
server option.
(Bug #13333431)
Replication: A race condition could occur when running multiple instances of mysqld on a single machine, when more than slave thread was started at the same time, and each such thread tried to use the same temporary file concurrently. (Bug #12844302, Bug #62055)
Replication:
It was possible on replication slaves where
FEDERATED
tables were in use to get
timeouts on long-running operations, such as Error 1160
Got an error writing communication
packets. The FEDERATED
tables did
not need to be replicated for the issue to occur.
(Bug #11758931, Bug #51196)
References: See also: Bug #12896628, Bug #61790.
Replication:
Statements that wrote to tables with
AUTO_INCREMENT
columns based on an unordered
SELECT
from another table could lead to the
master and the slave going out of sync, as the order in which
the rows are retrieved from the table may differ between them.
Such statements include any
INSERT ...
SELECT
,
REPLACE ...
SELECT
, or
CREATE
TABLE ... SELECT
statement. Such statements are now
marked as unsafe for statement-based replication, which causes
the execution of one to throw a warning, and forces the
statement to be logged using the row-based format if the logging
format is MIXED
.
(Bug #11758263, Bug #50440)
Replication:
On Windows replication slave hosts, STOP
SLAVE
took an excessive length of time to complete
when the master was down.
(Bug #11752315, Bug #43460)
Replication:
mysqlbinlog
--database=
included all dbname
SET
INSERT_ID=
assignments
from the binary log in its output, even if database
n
dbname
was never referenced in the
binary log. This was due to the fact that
COMMIT
statements were not
associated with any database in the binary log. Now in such
cases, the current database is tracked so that only those
SET INSERT_ID
assignments that are made in
the context of changes to tables in database
dbname
are actually printed in the
mysqlbinlog output.
(Bug #11746146, Bug #23894)
References: See also: Bug #23890, Bug #46998, Bug #11761686, Bug #54201, Bug #11754117, Bug #45670.
Microsoft Windows: On Windows, rebuilds in a source distribution failed to create the initial database due to insufficient cleanup from the previous run or failure to find the proper server executable. (Bug #13431251)
Microsoft Windows:
On Windows, the server incorrectly constructed the full path
name of the plugin binary for INSTALL
PLUGIN
and
CREATE
FUNCTION ... SONAME
.
(Bug #45549, Bug #11754014)
mysqldump tried to execute
SET
statements as SET OPTION
, which failed when
used against 5.6 or higher servers because the deprecated
OPTION
keyword has been removed from
SET
syntax.
(Bug #13813473)
The optimizer did not perform constant propagation for views, so a query containing views resulted in a less efficient execution plan than the corresponding query using only base tables. (Bug #13783777)
A memory leak could occur for queries containing a subquery that
used GROUP BY
on an outer column.
(Bug #13724099)
After using an ALTER TABLE
statement to change the KEY_BLOCK_SIZE
property for an InnoDB
table, for example
when switching from an uncompressed to a compressed table,
subsequent server restarts could fail with a message like:
InnoDB: Error: data file path
/ibdata2 uses page size 1024,
InnoDB: but the only supported page size in this release is=16384
This issue is a regression introduced in MySQL 5.5.20. (Bug #13698765, Bug #64160)
In debug builds, a Debug Sync timeout warning was treated as an error, causing an assertion to be raised. (Bug #13688248)
_mi_print_key()
iterated one time too many
when there was a NULL
bit, resulting in
Valgrind warnings.
(Bug #13686970)
Pushing down to InnoDB
an index
condition that called a stored function resulted in a server
crash. This kind of condition is no longer pushed down.
(Bug #13655397)
A SELECT
from a subquery that
returned an empty result could itself fail to return an empty
result as expected.
(Bug #13651009, Bug #13650418)
For debug builds, negative values with a zero integer part and nonzero fractional part (such as -0.1111) were not detected, so the negative fractional part was later cast to a large unsigned number and raised an assertion. (Bug #13616434)
If during server startup a signal such as
SIGHUP
was caught prior to full server
initialization, the server could crash. This was due to a race
condition between the signal handler thread and the main thread
performing server initialization. To prevent this from
happening, signal processing is now suspended until full
initialization of all server subsystems has been completed
successfully.
(Bug #13608371, Bug #62311)
The shared version of libmysqlclient
did not
export these functions for linking by client programs:
get_tty_password()
,
handle_options()
,
my_print_help()
.
(Bug #13604121)
An aggregated expression of type MIN()
or
MAX()
should return NULL
but could instead return the empty set if the query was
implicitly grouped and there was no HAVING
clause that evaluates to FALSE
.
(Bug #13599013)
Left join queries could be incorrectly converted to inner joins and return erroneous result sets. (Bug #13595212)
Date-handling code could raise an assertion attempting to calculate the number of seconds since the epoch. (Bug #13545236)
For queries that used a join type of
ref_or_null
, the optimizer
could skip the filesort operation and sort the results
incorrectly.
(Bug #13531865)
For some queries, a filesort operation was done even when the result contained only a single row and needed no sorting. (Bug #13529048)
The optimizer could return an incorrect select limit in some
cases when a query included no explicit LIMIT
clause.
(Bug #13528826)
In some cases, the optimizer failed to use a covering index when that was possible and read data rows instead. (Bug #13514959)
SELECT
statements failed for the
EXAMPLE
storage engine.
(Bug #13511529)
References: This issue is a regression of: Bug #11746275.
The Performance Schema instrumentation for stages did not fully
honor the ENABLED
column in the
schema.setup_instruments
table.
(Bug #13509513)
Converting a string ending with a decimal point (such as
'1.'
) to a floating-point number raised a
data truncation warning.
(Bug #13500371)
Use of an uninitialized TABLE_SHARE
member
could cause a server crash.
(Bug #13489996)
Some outer joins that used views as inner tables did not evaluate conditions correctly. (Bug #13464334)
A query that used an index on a
CHAR
column referenced in a
BETWEEN
clause could return invalid results.
(Bug #13463488, Bug #63437)
Expressions that compared a
BIGINT
column with any
non-integer constant were performed using integers rather than
decimal or float values, with the result that the constant could
be truncated. This could lead to any such comparison that used
<
,
>
,
<=
,
>=
,
=
,
!=
/<>
,
IN
, or
BETWEEN
yielding false positive or
negative results.
(Bug #13463415, Bug #11758543, Bug #63502, Bug #50756)
An application linked against libmysqld
could
crash in debug mode with a stack smashing
detected
error if it tried to connect without
specifying the user name.
(Bug #13460909)
Instantiating a derived table for a query with an empty result caused a server crash. (Bug #13457552)
When the optimizer performed conversion of
DECIMAL
values while evaluating
range conditions, it could produce incorrect results.
(Bug #13453382)
Implicitly grouped queries with a const
table
and no matching rows could return incorrect results.
(Bug #13430588)
For debug builds, enabling
optimizer_trace
could cause an
assertion to be raised.
(Bug #13430443)
Enabling index condition pushdown could cause performance degradation. (Bug #13430436)
When a fixed-width row was inserted into a
MyISAM
temporary table, the entire
content of the record buffer was written to the table, including
any trailing space contained in
VARCHAR
columns, the issue being
that this trailing space could be uninitialized. This problem
has been resolved by insuring that only the bytes actually used
to store the VARCHAR
(and none extra) are
copied and inserted in such cases.
(Bug #13389854, Bug #79028, Bug #22123583)
Fractional seconds parts were lost for certain
UNION ALL
queries.
(Bug #13375823)
When merging ranges that effectively resulted in a full index scan, the optimizer did not discard the range predicate as unneeded. (Bug #13354910)
When executing EXPLAIN
, it was
assumed that only the default multi-range read implementation
could produce an ordered result; this meant that when a query on
a table that used a storage engine providing its own sorted MRR,
it was ignored, so that EXPLAIN
failed to
report Using MRR
even when a multi-range read
was used.
(Bug #13330645)
Some multiple-table updates could update a row twice. (Bug #13095459)
Performance Schema idle
event timings were
not normalized to the same units as wait timings.
(Bug #13018537)
In MySQL 5.6.3, a number of status variables were changed to
longlong
types so that they would roll over
much later. However, the format string used by
mysqladmin status to print Queries
per second
values did not reflect this, causing such
values to be misreported.
(Bug #12990746)
References: See also: Bug #42698. This issue is a regression of: Bug #11751727.
For debug builds, two assertions could be raised erroneously for
UPDATE
statements.
(Bug #12912171)
When the result of a stored function returning a non-integer
type was evaluated for NULL
, an incorrect
type warning (Warning 1292 Truncated incorrect
INTEGER value) is generated, although such a test
for NULL should work with any type. This
could cause stored routines not handling the warning correctly
to fail.
The issue could be worked around by wrapping the result in an
expression, using a function such as
CONCAT()
.
(Bug #12872824, Bug #62125)
When running mysqldump with both the
--single-transaction
and
--flush-logs
options, the
flushing of the log performed an implicit
COMMIT
(see
Statements That Cause an Implicit Commit), causing more than one
transaction to be used and thus breaking consistency.
(Bug #12809202, Bug #61854)
A query that used an aggregate function such as
MAX()
or
MIN()
of an index with
NOT BETWEEN
in the WHERE
clause could fail to match rows, thus returning an invalid
result.
(Bug #12773464, Bug #61925)
With ONLY_FULL_GROUP_BY
SQL
mode enabled, columns that were not aggregated in the select
list or named in a GROUP BY
were incorrectly
permitted in ORDER BY
.
(Bug #12626418)
Mishandling of
NO_BACKSLASH_ESCAPES
SQL mode
within stored procedures on slave servers could cause
replication failures.
(Bug #12601974)
Passing a user variable as an argument to
GROUP_CONCAT()
could cause a
server exit if the variable value changed during query
execution.
(Bug #12408412)
LOAD INDEX INTO
CACHE
could cause a server exit if the index cache was
too small.
(Bug #12361113)
With ONLY_FULL_GROUP_BY
SQL mode enabled, a
query that uses GROUP BY
on a column derived
from a subquery in the FROM
clause failed
with a column isn't in GROUP BY
error, if the
query was in a view.
(Bug #11923239)
Attempting to execute ALTER TABLE
on a temporary MERGE
table having
an underlying temporary table rendered the
MERGE
table unusable, unless the
ALTER TABLE
specified a new list of
underlying tables.
(Bug #11764786, Bug #57657)
It was possible in the event of successive failures for mysqld_safe to restart quickly enough to consume excessive amounts of CPU. Now, on systems that support the sleep and date system utilities, mysqld_safe checks to see whether it has restarted more than 5 times in the current second, and if so, waits 1 second before attempting another restart. (Bug #11761530, Bug #54035)
A HAVING
clause in a query using
MIN()
or
MAX()
was sometimes ignored.
(Bug #11760517, Bug #52935)
References: See also: Bug #11758970, Bug #51242, Bug #11759718, Bug #52051.
When used with the --xml
option, mysqldump
--routines
failed to dump any
stored routines, triggers, or events.
(Bug #11760384, Bug #52792)
If an attempt to initiate a statement failed, the issue could not be reported to the client because it was not prepared to receive any error messages prior to the execution of any statement. Since the user could not execute any queries, they were simply disconnected without providing a clear error.
After the fix for this issue, the client is prepared for an error as soon as it attempts to initiate a statement, so that the error can be reported prior to disconnecting the user. (Bug #11755281, Bug #47032)
Previously, .OLD
files were not included
among the files deleted by DROP
DATABASE
. Files with this extension are now also
deleted by the statement.
(Bug #11751736, Bug #42708)
A prepared statement using a view whose definition changed between preparation and execution continued to use the old definition, which could cause the prepared statement to return incorrect results. (Bug #11748352, Bug #36002)
Some debugging information was written to the buffer after a flush, resulting in the information not appearing until the next flush. (Bug #64048, Bug #13608112)
Locale information for FORMAT()
function instances was lost in view definitions.
(Bug #63020, Bug #13344643)
mysqlhotcopy failed for databases containing views. (Bug #62472, Bug #13006947, Bug #12992993)
The VIO description string was initialized even for connections where it was unneeded. (Bug #62285, Bug #12951586)
On Windows, pasting multiple-line input including a CRLF terminator on the last line into the mysql client resulted in the first character of the last line being changed, resulting in erroneous statements. Handling of newlines in pasted input was also incorrect. (Bug #60901, Bug #12589167, Bug #64104, Bug #13639107)
The contents of the shared
and
shared-compat
RPM packages had been changed
in versions 5.5.6 and 5.6.1 to avoid the overlap which they
traditionally had (and still have in MySQL 5.0 and 5.1).
However, the RPM meta information had not been changed in
accordance, and so RPM still assumed a conflict between
shared
and shared-compat
RPM packages. This has been fixed.
(Bug #60855, Bug #12368215)
References: See also: Bug #56150.
The result of SUBSTRING_INDEX()
could be missing characters when used as an argument to
conversion functions such as
LOWER()
.
(Bug #60166, Bug #11829861)
UPDATE IGNORE
returned an incorrect count for
number of rows updated when there were duplicate-key conflicts
in a multiple-table update.
(Bug #59715, Bug #11766576)
The optimizer mishandled STRAIGHT_JOIN
used
with nested joins; for example, by not evaluating tables in the
specified order.
(Bug #59487, Bug #11766384, Bug #43368, Bug #11752239, Bug #60080, Bug #11766858)
A subquery involved in a comparison requiring a character set conversion caused an error that resulted in a server crash. (Bug #59185, Bug #11766143)
The embedded server crashed when argc = 0
.
(Bug #57931, Bug #12561297)
If tables were locked by
LOCK TABLES ...
READ
in another session, SET GLOBAL read_only
= 1
failed to complete.
(Bug #57612, Bug #11764747)
Invalid memory reads could occur when
cmp_item_sort_string::store_value()
tried to
refer to a temporary value that could be changed or deleted by
other functions.
(Bug #57510, Bug #11764651)
Assigning the result of a subquery to a user variable raised an
assertion when the outer query included
DISTINCT
and GROUP BY
.
(Bug #57196, Bug #11764371)
For comparisons containing out-of-range constants, the optimizer permitted warnings to leak through to the client, even though it accounted for the range issue internally. (Bug #56962, Bug #11764155)
A confusing CREATE TABLE
error
message was improved.
(Bug #54963, Bug #11762377)
The handle_segfault()
signal-handler code in
mysqld could itself crash due to calling
unsafe functions.
(Bug #54082, Bug #11761576)
Using myisamchk with the sort recover method to repair a table having fixed-width row format could cause the row pointer size to be reduced, effectively resulting in a smaller maximum data file size. (Bug #48848, Bug #11756869)
Enabling myisam_use_mmap
could
cause the server to crash.
(Bug #48726, Bug #11756764)
For MEMORY
tables, a scan of a
HASH
index on a
VARCHAR
column could fail to find
some rows if the index was on a prefix of the column.
(Bug #47704, Bug #11755870)
myisam_sort_buffer_size
could
not be set larger than 4GB on 64-bit systems.
(Bug #45702, Bug #11754145)
The stored routine cache was subject to a small memory leak that over time or with many routines being used could result in out-of-memory errors.
The fix for this issue also introduces a new global server
system variable
stored_program_cache
which can
be used for controlling the size of the stored routine cache.
(Bug #44585, Bug #11753187)
Under some circumstances, the result of
SUBSTRING_INDEX()
incorrectly
depended on the contents of the previous row.
(Bug #42404, Bug #11751514)
Setting an event to DISABLED
status and with
the ON COMPLETION NOT PRESERVE
attribute
caused it to be dropped at the next server restart.
(Bug #37666, Bug #11748899)
Due to improper locking, concurrent inserts into an
ARCHIVE
table at the same time as
repair and check operations on the table resulted in table
corruption.
(Bug #37280, Bug #11748748)
Stored functions could produce an error message that referred to
ORDER BY
even though the offending statement
within the function had no such clause.
(Bug #35410, Bug #11748187)