General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
variable. The problem was that THD::connect_utime could be
used without being initialized when the main thread is used
to handle connections (--thread-handling=no-threads).
The problem is that when a optimization of read-only transactions
(bypass 2-phase commit) was implemented, it removed the code that
reseted the XID once a transaction wasn't active anymore:
sql/sql_parse.cc:
- bzero(&thd->transaction.stmt, sizeof(thd->transaction.stmt));
- if (!thd->active_transaction())
- thd->transaction.xid_state.xid.null();
+ thd->transaction.stmt.reset();
This mostly worked fine as the transaction commit and rollback
functions (in handler.cc) reset the XID once the transaction is
ended. But those functions wouldn't reset the XID in case of
a empty transaction, leading to a assertion when a new starting
a new XA transaction.
The solution is to ensure that the XID state is reset when empty
transactions are ended (by either commit or rollback). This is
achieved by reorganizing the code so that the transaction cleanup
routine is invoked whenever a transaction is ended.
Make the caller of Query_log_event, Execute_load_log_event
constructors and THD::binlog_query to provide the error code
instead of having the constructors to figure out the error code.
MySQL crashes if a user without proper privileges attempts to create a procedure.
The crash happens because more than one error state is pushed onto the Diagnostic
area. In this particular case the user is denied to implicitly create a new user
account with the implicitly granted privileges ALTER- and EXECUTE ROUTINE.
The new account is needed if the original user account contained a host mask.
A user account with a host mask is a distinct user account in this context.
An alternative would be to first get the most permissive user account which
include the current user connection and then assign privileges to that
account. This behavior change is considered out of scope for this bug patch.
The implicit assignment of privileges when a user creates a stored routine is a
considered to be a feature for user convenience and as such it is not
a critical operation. Any failure to complete this operation is thus considered
non-fatal (an error becomes a warning).
The patch back ports a stack implementation of the internal error handler interface.
This enables the use of multiple error handlers so that it is possible to intercept
and cancel errors thrown by lower layers. This is needed as a error handler already
is used in the call stack emitting the errors which needs to be converted.
due to name_const substitution
Problem:
"In general, statements executed within a stored procedure
are written to the binary log using the same rules that
would apply were the statements to be executed in standalone
fashion. Some special care is taken when logging procedure
statements because statement execution within procedures
is not quite the same as in non-procedure context".
For example, each reference to a local variable in SP's
statements is replaced by NAME_CONST(var_name, var_value).
Queries like
"CREATE TABLE ... SELECT FUNC(local_var ..."
are logged as
"CREATE TABLE ... SELECT FUNC(NAME_CONST("local_var", var_value) ..."
that leads to differrent field names and
might result in "Incorrect column name" if var_value is long enough.
Fix: in 5.x we'll issue a warning in such a case.
In 6.0 we should get rid of NAME_CONST().
Note: this issue and change should be described in the documentation
("Binary Logging of Stored Programs").
Fine-tuning. Broke out comparison into method by
suggestion of Davi. Clarified comments. Reverting
test-case which I find too brittle; proper test
case in 5.1+.
When binlog_format is STATEMENT and the statement is unsafe before,
the unsafe warning/error message was issued without checking
whether the SQL_LOG_BIN was turned on or not.
Fixed with adding a sql_log_bin_toplevel flag in THD to check
whether SQL_LOG_BIN is ON in current session whatever the current is in sp or not.
- Add support for setting it as a server commandline argument
- Add support for those switches:
= no_index_merge
= no_index_merge_union
= no_index_merge_sort_union
= no_index_merge_intersection
- Remove bothersome warning messages. This change focuses on the warnings
that are covered by the ignore file: support-files/compiler_warnings.supp.
- Strings are guaranteed to be max uint in length
conflicts:
Text conflict in client/mysqltest.cc
Text conflict in mysql-test/include/wait_until_connected_again.inc
Text conflict in mysql-test/lib/mtr_report.pm
Text conflict in mysql-test/mysql-test-run.pl
Text conflict in mysql-test/r/events_bugs.result
Text conflict in mysql-test/r/log_state.result
Text conflict in mysql-test/r/myisam_data_pointer_size_func.result
Text conflict in mysql-test/r/mysqlcheck.result
Text conflict in mysql-test/r/query_cache.result
Text conflict in mysql-test/r/status.result
Text conflict in mysql-test/suite/binlog/r/binlog_index.result
Text conflict in mysql-test/suite/binlog/r/binlog_innodb.result
Text conflict in mysql-test/suite/rpl/r/rpl_packet.result
Text conflict in mysql-test/suite/rpl/t/rpl_packet.test
Text conflict in mysql-test/t/disabled.def
Text conflict in mysql-test/t/events_bugs.test
Text conflict in mysql-test/t/log_state.test
Text conflict in mysql-test/t/myisam_data_pointer_size_func.test
Text conflict in mysql-test/t/mysqlcheck.test
Text conflict in mysql-test/t/query_cache.test
Text conflict in mysql-test/t/rpl_init_slave_func.test
Text conflict in mysql-test/t/status.test
BUG#39325 Server crash inside MYSQL_LOG::purge_first_log halts replicaiton
The patch reverses the order of the purging and updating events for log and relay-log.info/index files respectively.
This solves the problem of having holes caused by crashes happening between updating info/index files and purging logs.
NOTE: This is a combined patch for BUG#38826 and BUG#39325. This patch is based on bugteam tree and takes into account reviewers suggestions.
The problem was that the server did not robustly handle a
unilateral roll back issued by the Resource Manager (RM)
due to a resource deadlock within the transaction branch.
By not acknowledging the roll back, the server (TM) would
eventually corrupt the XA transaction state and crash.
The solution is to mark the transaction as rollback-only
if the RM indicates that it rolled back its branch of the
transaction.
The problem was that the server did not robustly handle a
unilateral roll back issued by the Resource Manager (RM)
due to a resource deadlock within the transaction branch.
By not acknowledging the roll back, the server (TM) would
eventually corrupt the XA transaction state and crash.
The solution is to mark the transaction as rollback-only
if the RM indicates that it rolled back its branch of the
transaction.
When running Stored Routines the Status Variable "Questions" was wrongly
incremented. According to the manual it should contain the "number of
statements that clients have sent to the server"
Introduced a new status variable 'questions' to replace the query_id
variable which currently corresponds badly with the number of statements
sent by the client.
The new behavior is ment to be backward compatible with 4.0 and at the
same time work with new features in a similar way.
This is a backport from 6.0
The failure was caused by executing a CREATE-SELECT statement that creates a
table in another database than the current one. In row-based logging, the
CREATE statement was written to the binary log without the database, hence
creating the table in the wrong database, causing the following inserts to
fail since the table didn't exist in the given database.
Fixed the bug by adding a parameter to store_create_info() that will make
the function print the database name before the table name and used that
in the calls that write the CREATE statement to the binary log. The database
name is only printed if it is different than the currently selected database.
The output of SHOW CREATE TABLE has not changed and is still printed without
the database name.
"Trigger fired multiple times leads to gaps in auto_increment sequence".
The bug was that if a trigger fired multiple times inside a top
statement (for example top-statement is a multi-row INSERT,
and trigger is ON INSERT), and that trigger inserted into an auto_increment
column, then gaps could be observed in the auto_increment sequence,
even if there were no other users of the database (no concurrency).
It was wrong usage of THD::auto_inc_intervals_in_cur_stmt_for_binlog.
Note that the fix changes "class handler", I'll tell the Storage Engine API team.
The assertion indicates that some data was left in the transaction
cache when the server was shut down, which means that a previous
statement did not commit or rollback correctly.
What happened was that a bug in the rollback of a transactional
table caused the transaction cache to be emptied, but not reset.
The error can be triggered by having a failing UPDATE or INSERT,
on a transactional table, causing an implicit rollback.
Fixed by always flushing the pending event to reset the state
properly.
The failure was caused by executing a CREATE-SELECT statement that creates a
table in another database than the current one. In row-based logging, the
CREATE statement was written to the binary log without the database, hence
creating the table in the wrong database, causing the following inserts to
fail since the table didn't exist in the given database.
Fixed the bug by adding a parameter to store_create_info() that will make
the function print the database name before the table name and used that
in the calls that write the CREATE statement to the binary log. The database
name is only printed if it is different than the currently selected database.
The output of SHOW CREATE TABLE has not changed and is still printed without
the database name.
The problem of this bug is that we need to get the list of tables
to be updated for a multi-table update statement, which requires to
open all the tables referenced by the statement and resolve all
the fields involved in update in order to figure out the list of
tables for update. However if there are replicate filter rules,
some tables might not exist on slave and result in a failure
before we could examine the filter rules.
I think the whole problem can not be solved on slave alone,
the master must record and send the information of tables
involved for update to slave, so that the slave do not need to
open all the tables referenced by the multi-table update statement to
figure out which tables are involved for update.
So a status variable is added to Query_log event to store the
value of table map for update on master. And on slave, it will
try to get the value of this variable and use it to examine
filter rules without opening any tables on slave, if this values
is not available, the old approach is used and thus the bug will
still occur for when replicating from old masters.
build)
The crash was caused by freeing the internal parser stack during the parser
execution.
This occured only for complex stored procedures, after reallocating the parser
stack using my_yyoverflow(), with the following C call stack:
- MYSQLparse()
- any rule calling sp_head::restore_lex()
- lex_end()
- x_free(lex->yacc_yyss), xfree(lex->yacc_yyvs)
The root cause is the implementation of stored procedures, which breaks the
assumption from 4.1 that there is only one LEX structure per parser call.
The solution is to separate the LEX structure into:
- attributes that represent a statement (the current LEX structure),
- attributes that relate to the syntax parser itself (Yacc_state),
so that parsing multiple statements in stored programs can create multiple
LEX structures while not changing the unique Yacc_state.
Now, Yacc_state and the existing Lex_input_stream are aggregated into
Parser_state, a structure that represent the complete state of the (Lexical +
Syntax) parser.
subselects into account
It is forbidden to use the SELECT INTO construction inside UNION statements
unless on the last SELECT of the union. The parser records whether it
has seen INTO or not when parsing a UNION statement. But if the INTO was
legally used in an outer query, an error is thrown if UNION is seen in a
subquery. Fixed in 5.0 by remembering the nesting level of INTO tokens and
mitigate the error unless it collides with the UNION.
