in short we now record whenever the slave I/O thread ignores a master's event because of its server id,
and use this info in the slave SQL thread to advance Exec_master_log_pos. Because if we
do not, this variable stays at the position of the last executed event, i.e. the last *non-ignored*
executed one, which may not be the last of the master's binlog (and so the slave *looks* behind
the master though it's data-wise it's not).
binlog-do-db or binlog-ignore-db are in effect. (In the future 5.1? 5.0? I think
each statement should be verified against the filtering criteria based on the database
it *uses* and not the *current* one. But, right now the *current* database is what
counts according to the semantics of the manual.)
WL#2335 (wait if binlog or binlog index file hits disk full or quota exceeded),
fix for BUG#7236 ("--expire_logs_days does not apply if all statements
happen in transactions"), and a behaviour change: abort if mysqld can't start
binlog at startup (if running with --log-bin of course).
tables requires privileges for them if some table or column level grants
present" (with after-review fixes).
We should set SELECT_ACL for implicitly opened tables in
my_tz_check_n_skip_implicit_tables() to be able to bypass privilege
checking in check_grant(). Also we should exclude those tables from
privilege checking in multi-update.
CREATE DATABASE statement used the current database instead of the
database created when checking conditions for replication.
CREATE/DROP/ALTER DATABASE statements are now replicated based on
the manipulated database.
When we are writing a transaction to the binlog, we log BEGIN/COMMIT with zero error code.
Example: all statements of trans succeeded, connection lost and so implicit rollback:
we don't want ER_NET* errors to be logged in the BEGIN/ROLLBACK events, while statement
events have 0. If there was really a serious error code, it's already in the statement events.
the fact that the transaction log is empty does not mean we're not in a transaction
(it could be BEGIN; SELECT * FOR UPDATE FROM ibtable: then we don't want to commit now, even if
the statement is a MyISAM update).
With a testcase.
We must not reset the charset in slave after each statement, otherwise the SET CHARACTER SET is cancelled immediately.
Instead, we write a SET CHARACTER SET DEFAULT to the master's binlog when needed (like we already do for SET FOREIGN_KEY_CHECKS);
such writing is not necessary in 4.1 (in 4.1 the bug does not exist, as the SET ONE_SHOT syntax is used).
I have written a test and it works, but I'm not pushing the test as it requires building with all charsets.
I have noticed differences between what is inserted in the master's table in 4.0 and 4.1, and alerted Bar.
Fix for BUG#4971 "CREATE TABLE ... TYPE=HEAP SELECT ... stops slave (wrong DELETE in binlog)":
replacing the no_log argument of mysql_create_table() by some safer method
(temporarily setting OPTION_BIN_LOG to 0) which guarantees that even the automatic
DELETE FROM heap_table does not get into the binlog when a not-yet-existing HEAP table
is opened by mysql_create_table().
mysqld.cc:
Changed LOGLEVEL enum to loglevel
mysql_priv.h, log.cc:
Changed LOGLEVEL to loglevel. Removed startup_ from some of the DBUG_ENTER macros. Removed the print_msg_to_log function as it was unused.
my_getopt.c, my_getopt.h:
Renamed LOGLEVEL to loglevel to match coding standards
replacing the no_log argument of mysql_create_table() by some safer method
(temporarily setting OPTION_BIN_LOG to 0) which guarantees that even the automatic
DELETE FROM heap_table does not get into the binlog when a not-yet-existing HEAP table
is opened by mysql_create_table().
mysqld.cc:
Changed option_error_reporter to match new function header that includes LOGLEVEL enum
mysql_priv.h:
Removed the MY_ERROR style bitmask. Changed function headers to use new LOGLEVEL enum
log.cc:
Changed print_buffer_to_log to print_buffer_to_file. Remove the timestamp bool and now all log entries written to stderr are timestamped. Removed some unused commented code. changed to use the new LOGLEVEL enum.
my_getopt.c:
Changed functions to use the new LOGLEVEL enum and changed the included error reporter to be default_reporter. This reporter is used in handle_options if a reporter is not given
my_getopt.h:
changed typedefs to use better naming convention. Moved error bitmask into the LOGLEVEL enum and included it here.
Added declarations for print_msg_to_log and vprint_msg_to_log. sql_print_error are simple functions that wrap calls to print_msg_to_log. Define the different error types with MY_ERROR_TYPE, MY_WARNING_TYPE, and MY_INFORMATION_TYPE
gen_lex_hash.cc:
Added NULL error reporting parameter to handle_options
log.cc:
Add print_msg_to_log, print_buffer_to_log, and vprint_msg_to_log. Print_msg_to_log will write the message to the windows event log if on NT. We now have error, warning, and information versions of sql_print_xxxx. T his is a variation of a similar changeset WAX did.
mysqld.cc:
Added option_error_reporter callback function and pass that into handle_options
mysql.cc:
Added NULL as error reporter arg to the end of handle_options
Many files:
Added NULL error reporter parameter as the last paramter to handle_options
my_getopt.c:
Added second function pointer to server as an error reporting callback. Added local function report_option_error that will either write the error to stderr or to the error reporting callback. changed all calls in handle_options from fprintf(stderr, ... ) to report_option_error
my_getopt.h:
Changed declaration of handle_options to use typedefs for the two function pointers. added second function pointer to server as an error reporting callback
mysqld.dsp:
Added custom build step for compiling message file and added message resource file (output of mc)
Bug #4810 "deadlock with KILL when the victim was in a wait state"
(I included mutex unlock into exit_cond() for future safety)
and BUG#4827 "KILL while START SLAVE may lead to replication slave crash"
happen on SMP machines, when a thread is going to wait on a condition and it is KILLed at the
same time.
Cleaning code a bit by adding a test in enter_cond() that we have the mutex (was already the case in all places
where it's called except one which is fixed here).
a limitation of yesterday's implementation:
if there was an unfinished transaction (COMMIT not typed), and some MyISAM tables were
then updated, and then mysqld crashes, then at restart the server would use the too old
binlog offset known by InnoDB to cut the binlog, thus cutting the successful MyISAM
updates. We fix this by reporting the binlog offset into InnoDB even if InnoDB was not
affected at all by the update.
But the feature is still disabled until we decide if it can go into 4.1.3.
Final push. Printing some warnings at startup, as --innodb-safe-binlog requires
some other options, to work as expected.
Adding 6 new tests (3 pairs). If they fail on some platforms (so far they have been run only on my Linux),
they should be ignored for the 4.1.3 build (you can just
rm mysql-test/*/rpl_crash_*).
Now going to update doc.
Won't be pushed as is - separate email sent for internal review.
WL#1717 "binlog-innodb consistency".
Now when mysqld starts, if InnoDB does a crash recovery, we use the binlog name
and position retrieved from InnoDB (corresponding to the last transaction
successfully committed by InnoDB) to cut any rolled back transaction from
the binary log. This is triggered by the --innodb-safe-binlog option.
Provided you configure mysqld to fsync() InnoDB at every commit (using
flush_log_at_trx_commit) and to fsync() the binlog at every write
(using --sync-binlog=1), this behaviour guarantees that a master always has
consistency between binlog and InnoDB, whenever the crash happens.
6 tests to verify that it works.
Added basic per-thread time zone functionality (based on public
domain elsie-code). Now user can select current time zone
(from the list of time zones described in system tables).
All NOW-like functions honor this time zone, values of TIMESTAMP
type are interpreted as values in this time zone, so now
our TIMESTAMP type behaves similar to Oracle's TIMESTAMP WITH
LOCAL TIME ZONE (or proper PostgresSQL type).
WL#1266 "CONVERT_TZ() - basic time with time zone conversion
function".
Fixed problems described in Bug #2336 (Different number of warnings
when inserting bad datetime as string or as number). This required
reworking of datetime realted warning hadling (they now generated
at Field object level not in conversion functions).
Optimization: Now Field class descendants use table->in_use member
instead of current_thd macro.
New option --sync-binlog=x (and global settable variable) which will fsync the binlog
after every x-th disk write to it. That is, if in autocommit mode, after every x-th statement
written to the binlog; if using transactions, after every x-th transaction written to the binlog.
x==0 means no fsync. x==1 is the slowest.
There is no test added for this, I have just checked that it works as --sync-binlog=1 dramatically
slows down mysqld.
Made sync-frm a global settable variable.
by binlogging some SET ONE_SHOT CHARACTER_SETetc,
which will be enough until we have it more compact and more complete in 5.0. With the present patch,
replication will work ok between 4.1.3 master and slaves, as long as:
- master and slave have the same GLOBAL.COLLATION_SERVER
- COLLATION_DATABASE and CHARACTER_SET_DATABASE are not used
- application does not use the fact that table is created with charset of the USEd db (BUG#2326).
all of which are not too hard to fulfill.
ONE_SHOT is reserved for internal use of mysqlbinlog|mysql and works only for charsets,
so we give error if used for non-charset vars.
Fix for BUG#3875 "mysqlbinlog produces wrong ouput if query uses
variables containing quotes" and BUG#3943 "Queries with non-ASCII literals are not replicated
properly after SET NAMES".
Detecting that master and slave have different global charsets or server ids.
