Changes in MySQL 5.6.1 (Not released, Milestone 5)

Note

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.)

• Performance Schema Notes

• Functionality Added or Changed

• Bugs Fixed

Performance Schema Notes

• The Performance Schema has these additions:

  • The setup_consumers table contents have changed. Previously, the table used a “flat” structure with a one-to-one correspondence between consumer name and destination table. This has been replaced with a hierarchy of consumer settings that enable progressively finer control of which destinations receive events. The previous xxx_summary_xxx consumers no longer exist. Instead, the Performance Schema maintains appropriate summaries automatically for the levels for which settings in the consumer hierarchy are enabled. For example, if only the top-level (global) consumer is enabled, only global summaries are maintained. Others, such as thread-level summaries, are not. See Pre-Filtering by Consumer. In addition, optimizations have been added to reduce Performance Schema overhead.

  • It is now possible to filter events by object using the new setup_objects table. Currently, this table can be used to selectively instrument tables, based on schema names and/or table names. See Pre-Filtering by Object. A new table, objects_summary_global_by_type, summarizes events for objects.

  • It is now possible to filter events by thread, and the Performance Schema collects more information for each thread. A new table, setup_actors, can be used to selectively instrument user connections, based on the user name and/or host name of each connecting session. The threads table, which contains a row for each active server thread, was extended with several new columns. With these additions, the information available in threads is like that available from the INFORMATION_SCHEMA.PROCESSLIST table or the output from SHOW PROCESSLIST. Thus, all three serve to provide information for thread-monitoring purposes. Use of threads differs from use of the other two thread information sources in these ways:

    • Access to threads does not require a mutex and has minimal impact on server performance. INFORMATION_SCHEMA.PROCESSLIST and SHOW PROCESSLIST have negative performance consequences because they require a mutex.

    • threads provides additional information for each thread, such as whether it is a foreground or background thread, and the location within the server associated with the thread.

    • threads provides information about background threads. This means that threads can be used to monitor activity the other thread information sources cannot.

    • You can control which threads are monitored by setting the INSTRUMENTED column or by using the setup_actors table.

    For these reasons, DBAs who perform server monitoring using INFORMATION_SCHEMA.PROCESSLIST or SHOW PROCESSLIST may wish to monitor using threads instead.

  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.

  For more information, see MySQL Performance Schema.

Functionality Added or Changed

• Incompatible Change: The following obsolete constructs have been removed. Where alternatives are shown, applications should be updated to use them.

  • The --log server option and the log system variable. Instead, use the --general_log option to enable the general query log and the --general_log_file=file_name option to set the general query log file name.

  • The --log-slow-queries server option and the log_slow_queries system variable. Instead, use the --slow_query_log option to enable the slow query log and the --slow_query_log_file=file_name option to set the slow query log file name.

  • The --one-thread server option. Use --thread_handling=no-threads instead.

  • The --skip-thread-priority server option.

  • The engine_condition_pushdown system variable. Use the engine_condition_pushdown flag of the optimizer_switch variable instead.

  • The have_csv, have_innodb, have_ndbcluster, and have_partitioning system variables. Use SHOW ENGINES instead.

  • The sql_big_tables system variable. Use big_tables instead.

  • The sql_low_priority_updates system variable. Use low_priority_updates instead.

  • The sql_max_join_size system variable. Use max_join_size instead.

  • The SLAVE START and SLAVE STOP statements. Use the START SLAVE and STOP SLAVE statements instead.

  • The ONE_SHOT modifier for the SET statement.

• Important Change; Replication: Replication filtering options such as --replicate-do-db, --replicate-rewrite-db, and --replicate-do-table were not consistent with one another in regard to case sensitivity. Now all --replicate-* options follow the same rules for case sensitivity applying to names of databases and tables elsewhere in the MySQL server, including the effects of the lower_case_table_names system variable. (Bug #51639, Bug #11759334)

• Important Change; Replication: Added the MASTER_RETRY_COUNT option to the CHANGE MASTER TO statement, and a corresponding Master_Retry_Count column to the output of SHOW SLAVE STATUS. The option sets the value shown in this column. MASTER_RETRY_COUNT is intended eventually to replace the older (and now deprecated) --master-retry-count server option, and is now the preferred method for setting the maximum number of times that the slave may attempt to reconnect after losing its connection to the master. (Bug #44209, Bug #11752887, Bug #44486, Bug #11753110)

• InnoDB: Setting innodb_read_ahead_threshold to 0 disables read-ahead. Prior to 5.6.1, a value of 0 would trigger a read-ahead upon reading the boundary page of a 64 page extent. (Bug #11763876, Bug #56646)

• InnoDB: InnoDB can now report the total size of the rollback segment, measured in pages. The value is reported through the information_schema.innodb_metrics table, using the counter trx_rseg_current_size. You enable and query the counter as follows:

  mysql> SET GLOBAL innodb_monitor_enable = 'trx_rseg_current_size';
  
  mysql> SELECT name, count, max_count, comment
      ->     FROM innodb_metrics WHERE name = 'trx_rseg_current_size';
  +-----------------------+-------+-----------+----------------------------------------+
  | name                  | count | max_count | comment                                |
  +-----------------------+-------+-----------+----------------------------------------+
  | trx_rseg_current_size |   346 |       346 | Current rollback segment size in pages |
  +-----------------------+-------+-----------+----------------------------------------+

  (Bug #57584)

• Replication: SHOW SLAVE STATUS now displays the actual number of retries for each connection attempt made by the I/O thread. (Bug #56416, Bug #11763675)

• Replication: Added the Slave_last_heartbeat status variable, which shows when a replication slave last received a heartbeat signal. The value is displayed using TIMESTAMP format. (Bug #45441)

• Replication: Timestamps have been added to the output of SHOW SLAVE STATUS to show when the most recent I/O and SQL thread errors occurred. The Last_IO_Error column is now prefixed with the timestamp for the most recent I/O error, and Last_SQL_Error shows the timestamp for the most recent SQL thread error. The timestamp values use the format YYMMDD HH:MM:SS in both of these columns. For more information, see SHOW SLAVE STATUS Syntax. (Bug #43535, Bug #11752361, Bug #64255, Bug #13726435)

• There is now a bind_address system variable containing the value of the --bind-address option. This enables the address to be accessed at runtime. (Bug #44355, Bug #11752999)

• “Unknown table” error messages that included only the table name now include the database name as well. (Bug #34750, Bug #11747993)

• Previously, EXPLAIN output for a large union truncated the UNION RESULT row at the end of the list as follows if the string became too large:

  <union1,2,3,4,...>

  To make it easier to understand the union boundaries, truncation now occurs in the middle of the string:

  <union1,2,3,...,9>

  (Bug #30597, Bug #11747073)

• The following items are deprecated and will be removed in a future MySQL release. Where alternatives are shown, applications should be updated to use them.

  • The thread_concurrency system variable.

  • The --language server option. Use the lc_messages_dir and lc_messages system variables instead.

  • The --master-retry-count server option. Use the MASTER_RETRY_COUNT option the CHANGE MASTER TO statement instead.

• Support for adding Unicode collations that are based on the Unicode Collation Algorithm (UCA) has been improved:

  • MySQL now recognizes a larger subset of the LDML syntax that is used to write collation descriptions. In many cases, it is possible to download a collation definition from the Unicode Common Locale Data Repository and paste the relevant part (that is, the part between the <rules> and </rules> tags) into the MySQL Index.xml file.

  • Character representation in LDML rules is more flexible. Any character can be written literally, not just basic Latin letters. For collations based on UCA 5.2.0, hexadecimal notation can be used for any character, not just BMP characters.

  • When problems are found while parsing Index.xml, better diagnostics are produced.

  • For collations that require tailoring rules, there is no longer a fixed size limit on the tailoring information.

  For more information, see LDML Syntax Supported in MySQL, and Diagnostics During Index.xml Parsing.

• TO_BASE64() and FROM_BASE64() functions are now available to perform encoding to and from base-64 strings.

• The Unicode implementation has been extended to include a utf16le character set, which corresponds to the UTF-16LE encoding of the Unicode character set. This is similar to utf16 (UTF-16) but is little-endian rather than big-endian.

  Two utf16le collations are available:

  • utf16le_general_ci: The default collation, case sensitive (similar to utf16_general_ci).

  • utf16le_bin: Case sensitive, with by-codepoint comparison that provides the same order as utf16_bin.

  There are some limitations on the use of utf16le. With the exception of the item regarding user-defined collations, these are the same as the limitations on ucs2, utf16, and utf32.

  • utf16le cannot be used as a client character set, which means that it also does not work for SET NAMES or SET CHARACTER SET.

  • It is not possible to use LOAD DATA INFILE to load data files that use utf16le.

  • FULLTEXT indexes cannot be created on a column that uses utf16le. However, you can perform IN BOOLEAN MODE searches on the column without an index.

  • The use of ENCRYPT() with utf16le is not recommended because the underlying system call expects a string terminated by a zero byte.

  • It is not possible to create user-defined UCA collations for utf16le because there is no utf16le_unicode_ci collation, which would serve as the basis for such collations.

• Changes to replication in MySQL 5.6 make mysqlbinlog output generated by the --base64-output=ALWAYS option unusable. ALWAYS is now an invalid value for this option. If the option is given without a value, the effect is now the same as --base64-output=AUTO rather than --base64-output=ALWAYS.

  References: See also: Bug #28760.

• Croatian collations were added for Unicode character sets: utf8_croatian_ci, ucs2_croatian_ci, utf8mb4_croatian_ci, utf16_croatian_ci, and utf32_croatian_ci. Thee collations have tailoring for Croatian letters: Č, Ć, Dž, Đ, Lj, Nj, Š, Ž. They are based on Unicode 4.0.

• Several changes were made to optimizer-related system variables:

  • The optimizer_switch system variable has new engine_condition_pushdown and index_condition_pushdown flags to control whether storage engine condition pushdown and index condition pushdown optimizations are used. The engine_condition_pushdown system variable now is deprecated. For information about condition pushdown, see Engine Condition Pushdown Optimization, and Index Condition Pushdown Optimization.

  • The optimizer_switch system variable has new mrr and mrr_cost_based flags to control use of the Multi-Range Read optimization. The optimizer_use_mrr system variable has been removed. For information about Multi-Range Read, see Multi-Range Read Optimization.

  • The join_cache_level system variable has been renamed to optimizer_join_cache_level. This enables a single SHOW VARIABLES LIKE 'optimizer%' statement to show more optimizer-related settings.

• The Block Nested-Loop (BNL) Join algorithm previously used only for inner joins has been extended and can be employed for outer join operations, including nested outer joins. For more information, see Block Nested-Loop and Batched Key Access Joins.

  In conjunction with this work, a new system variable, optimizer_join_cache_level, controls how join buffering is done.

• The OpenGIS specification defines functions that test the relationship between two geometry values. MySQL originally implemented these functions such that they used object bounding rectangles and returned the same result as the corresponding MBR-based functions. Corresponding versions are now available that use precise object shapes. These versions are named with an ST_ prefix. For example, Contains() uses object bounding rectangles, whereas ST_Contains() uses object shapes. For more information, see Functions That Test Spatial Relations Between Geometry Objects.

  There are also now ST_ aliases for existing spatial functions that were already exact. For example, ST_IsEmpty() is an alias for IsEmpty()

  In addition, the IsSimple() and ST_Distance() spatial functions are now implemented, as well as the set operator functions ST_Difference(), ST_Intersection(), ST_SymDifference(), and ST_Union(), (Bug #4249, Bug #11744883)

• A --bind-address option has been added to a number of MySQL client programs: mysql, mysqldump, mysqladmin, mysqlbinlog, mysqlcheck, mysqlimport, and mysqlshow. This is for use on a computer having multiple network interfaces, and enables you to choose which interface is used to connect to the MySQL server.

  A corresponding change was made to the mysql_options() C API function, which now has a MYSQL_OPT_BIND option for specifying the interface. The argument is a host name or IP address (specified as a string).

Bugs Fixed

• Incompatible Change; Replication: The behavior of INSERT DELAYED statements when using statement-based replication has changed as follows:

  Previously, when using binlog_format=STATEMENT, a warning was issued in the client when executing INSERT DELAYED; now, no warning is issued in such cases.

  Previously, when using binlog_format=STATEMENT, INSERT DELAYED was logged as INSERT DELAYED; now, it is logged as an INSERT, without the DELAYED option.

  However, when binlog_format=STATEMENT, INSERT DELAYED continues to be executed as INSERT (without the DELAYED option). The behavior of INSERT DELAYED remains unchanged when using binlog_format=ROW: INSERT DELAYED generates no warnings, is executed as INSERT DELAYED, and is logged using the row-based format.

  This change also affects binlog_format=MIXED, because INSERT DELAYED is no longer considered unsafe. Now, when the logging format is MIXED, no switch to row-based logging occurs. This means that the statement is logged as a simple INSERT (that is, without the DELAYED option), using the statement-based logging format. (Bug #54579, Bug #11762035)

  References: See also: Bug #56678, Bug #11763907, Bug #57666. This issue is a regression of: Bug #39934, Bug #11749859.

• Incompatible Change; Replication: When determining whether to replicate a CREATE DATABASE, DROP DATABASE, or ALTER DATABASE statement, database-level options now take precedence over any --replicate-wild-do-table options. In other words, when trying to replicate one of these statements, --replicate-wild-do-table options are now checked if and only if there are no database-level options that apply to the statement. (Bug #46110, Bug #11754498)

• Incompatible Change: Starvation of FLUSH TABLES WITH READ LOCK statements occurred when there was a constant load of concurrent DML statements in two or more connections. Deadlock occurred when a connection that had some table open through a HANDLER statement tried to update data through a DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently.

  These problems resulted from the global read lock implementation, which was reimplemented with the following consequences:

  • To solve deadlock in event-handling code that was exposed by this patch, the LOCK_event_metadata mutex was replaced with metadata locks on events. As a result, DDL operations on events are now prohibited under LOCK TABLES. This is an incompatible change.

  • The global read lock (FLUSH TABLES WITH READ LOCK) no longer blocks DML and DDL on temporary tables. Before this patch, server behavior was not consistent in this respect: In some cases, DML/DDL statements on temporary tables were blocked; in others, they were not. Since the main use cases for FLUSH TABLES WITH READ LOCK are various forms of backups and temporary tables are not preserved during backups, the server now consistently permits DML/DDL on temporary tables under the global read lock.

  • The set of thread states has changed:

    • Waiting for global metadata lock is replaced by Waiting for global read lock.

    • Previously, Waiting for release of readlock was used to indicate that DML/DDL statements were waiting for release of a read lock and Waiting to get readlock was used to indicate that FLUSH TABLES WITH READ LOCK was waiting to acquire a global read lock. Now Waiting for global read lock is used for both cases.

    • Previously, Waiting for release of readlock was used for all statements that caused an explicit or implicit commit to indicate that they were waiting for release of a read lock and Waiting for all running commits to finish was used by FLUSH TABLES WITH READ LOCK. Now Waiting for commit lock is used for both cases.

    • There are two other new states, Waiting for trigger metadata lock and Waiting for event metadata lock.

  (Bug #57006, Bug #11764195, Bug #54673, Bug #11762116)

• Incompatible Change: CREATE TABLE statements (including CREATE TABLE ... LIKE) are now prohibited whenever a LOCK TABLES statement is in effect.

  One consequence of this change is that CREATE TABLE ... LIKE makes the same checks as CREATE TABLE and does not just copy the .frm file. This means that if the current SQL mode is different from the mode in effect when the original table was created, the table definition might be considered invalid for the new mode and the statement will fail. (Bug #42546, Bug #11751609)

• InnoDB; Replication: If the master had innodb_file_per_table=OFF, innodb_file_format=Antelope (and innodb_strict_mode=OFF), or both, certain CREATE TABLE options, such as KEY_BLOCK_SIZE, were ignored. This could permit the master to avoid raising ER_TOO_BIG_ROWSIZE errors.

  However, the ignored CREATE TABLE options were still written into the binary log, so that, if the slave had innodb_file_per_table=ON and innodb_file_format=Barracuda, it could encounter an ER_TOO_BIG_ROWSIZE error while executing the record from the log, causing the slave SQL thread to abort and replication to fail.

  In the case where the master was running MySQL 5.1 and the slave was MySQL 5.5 (or later), the failure occurred when both master and slave were running with default values for innodb_file_per_table and innodb_file_format. This could cause problems during upgrades.

  To address this issue, the default values for innodb_file_per_table and innodb_file_format are reverted to the MySQL 5.1 default values—that is, OFF and Antelope, respectively. (Bug #56318, Bug #11763590)

• InnoDB: If the MySQL Server crashed immediately after creating an InnoDB table, the server could quit with a signal 11 during the subsequent restart. The issue could occur if the server halted after InnoDB created the primary index for the table, but before the index definition was recorded in the MySQL metadata. (Bug #57616)

  References: This issue is a regression of: Bug #54582.

• InnoDB: With binary logging enabled, InnoDB could halt during crash recovery with a message referring to a transaction ID of 0. (Bug #54901, Bug #11762323)

• Replication: Due to changes made in MySQL 5.5.3, settings made in the binlog_cache_size and max_binlog_cache_size server system variables affected both the binary log statement cache (also introduced in that version) and the binary log transactional cache (formerly known simply as the binary log cache). This meant that the resources used as a result of setting either or both of these variables were double the amount expected. To rectify this problem, these variables now affect only the transactional cache. The fix for this issue also introduces two new system variables binlog_stmt_cache_size and max_binlog_stmt_cache_size, which affect only the binary log statement cache.

  In addition, the Binlog_cache_use status variable was incremented whenever either cache was used, and Binlog_cache_disk_use was incremented whenever the disk space from either cache was used, which caused problems with performance tuning of the statement and transactional caches, because it was not possible to determine which of these was being exceeded when attempting to troubleshoot excessive disk seeks and related problems. This issue is solved by changing the behavior of these two status variables such that they are incremented only in response to usage of the binary log transactional cache, as well as by introducing two new status variables Binlog_stmt_cache_use and Binlog_stmt_cache_disk_use, which are incremented only by usage of the binary log statement cache.

  The behavior of the max_binlog_cache_size system variable with regard to active sessions has also been changed to match that of the binlog_cache_size system variable: Previously, a change in max_binlog_cache_size took effect in existing sessions; now, as with a change in binlog_cache_size, a change in max_binlog_cache_size takes effect only in sessions begun after the value was changed.

  For more information, see System Variables Used with Binary Logging, and Server Status Variables. (Bug #57275, Bug #11764443)

• Replication: The Binlog_cache_use and Binlog_cache_disk_use status variables were incremented twice by a change to a table using a transactional storage engine. (Bug #56343, Bug #11763611)

  References: This issue is a regression of: Bug #50038.

• Replication: When STOP SLAVE is issued, the slave SQL thread rolls back the current transaction and stops immediately if the transaction updates only tables which use transactional storage engines. Previously, this occurred even when the transaction contained CREATE TEMPORARY TABLE statements, DROP TEMPORARY TABLE statements, or both, although these statements cannot be rolled back. Because temporary tables persist for the lifetime of a user session (in the case, the replication user), they remain until the slave is stopped or reset. When the transaction is restarted following a subsequent START SLAVE statement, the SQL thread aborts with an error that a temporary table to be created (or dropped) already exists (or does not exist, in the latter case).

  Following this fix, if an ongoing transaction contains CREATE TEMPORARY TABLE statements, DROP TEMPORARY TABLE statements, or both, the SQL thread now waits until the transaction ends, then stops. (Bug #56118, Bug #11763416)

• Replication: When an error occurred in the generation of the name for a new binary log file, the error was logged but not shown to the user. (Bug #46166)

  References: See also: Bug #37148, Bug #11748696, Bug #40611, Bug #11750196, Bug #43929, Bug #51019.

• Replication: When lower_case_table_names was set to 1 on the slave, but not on the master, names of databases in replicated statements were not converted, causing replication to fail on slaves using case-sensitive file systems. This occurred for both statement-based and row-based replication.

  In addition, when using row-based replication with lower_case_table_names set to 1 on the slave only, names of tables were also not converted, also causing replication failure on slaves using case-sensitive file systems. (Bug #37656)

• After setting collation_connection to one of the collations for the ucs2 or utf16 character sets, it was not possible to change the collation thereafter. (Bug #65000, Bug #13970475)

• cmake -DBUILD_CONFIG=mysql_release on Linux previously required libaio to be linked in. Now it is possible to specify -DIGNORE_AIO_CHECK to build without libaio. (Bug #58955, Bug #11765940)

• A Valgrind failure occurred in fn_format when called from archive_discover. (Bug #58205, Bug #11765259)

• Passing a string that was not null-terminated to UpdateXML() or ExtractValue() caused the server to fail with an assertion. (Bug #57279, Bug #11764447)

• In bootstrap mode, the server could not execute statements longer than 10,000 characters. (Bug #55817, Bug #11763139)

• NULL values were not grouped properly for some joins containing GROUP BY. (Bug #45267, Bug #11753766)

• A HAVING clause could be lost if an index for ORDER BY was available, incorrectly permitting additional rows to be returned. (Bug #45227, Bug #11753730)

• The optimizer could underestimate the memory required for column descriptors during join processing and cause memory corruption or a server crash. (Bug #42744, Bug #11751763)

• The server returned incorrect results for WHERE ... OR ... GROUP BY queries against InnoDB tables. (Bug #37977, Bug #11749031)

• An incorrectly checked XOR subquery optimization resulted in an assertion failure. (Bug #37899, Bug #11748998)

• A query that could use one index to produce the desired ordering and another index for range access with index condition pushdown could cause a server crash. (Bug #37851, Bug #11748981)

• With index condition pushdown enabled, InnoDB could crash due to a mismatch between what pushdown code expected to be in a record versus what was actually there. (Bug #36981, Bug #11748647)

• The range optimizer ignored conditions on inner tables in semi-join IN subqueries, causing the optimizer to miss good query execution plans. (Bug #35674, Bug #11748263)

• A server crash or memory overrun could occur with a dependent subquery and joins. (Bug #34799, Bug #11748009)

• Selecting from a view that referenced the same table in the FROM clause and an IN clause caused a server crash. (Bug #33245)

• Deeply nested subqueries could cause stack overflow or a server crash. (Bug #32680, Bug #11747503)

• The server crashed on optimization of queries that compared an indexed DECIMAL column with a string value. (Bug #32262, Bug #11747426)

• The server crashed on optimizations that used the range checked for each record access method. (Bug #32229, Bug #11747417)

• Contains() failed for multipolygon geometries. (Bug #32032, Bug #11747370)

• If the optimizer used a Multi-Range Read access method for index lookups, incorrect results could occur for rows that contained any BLOB or TEXT data types. (Bug #30622, Bug #11747076)

• Compared to MySQL 5.1, the optimizer failed to use join buffering for certain queries, resulting in slower performance for those queries. (Bug #30363, Bug #11747028)

• For Multi-Range Read scans used to resolve LIMIT queries, failure to close the scan caused file descriptor leaks for MyISAM tables. (Bug #30221, Bug #11746994)

• SHOW CREATE DATABASE did not account for the value of the lower_case_table_names system variable. (Bug #21317, Bug #11745926)