==== Description ====
Flashback can rollback the instances/databases/tables to an old snapshot.
It's implement on Server-Level by full image format binary logs (--binlog-row-image=FULL), so it supports all engines.
Currently, it’s a feature inside mysqlbinlog tool (with --flashback arguments).
Because the flashback binlog events will store in the memory, you should check if there is enough memory in your machine.
==== New Arguments to mysqlbinlog ====
--flashback (-B)
It will let mysqlbinlog to work on FLASHBACK mode.
==== New Arguments to mysqld ====
--flashback
Setup the server to use flashback. This enables binary log in row mode
and will enable extra logging for DDL's needed by flashback feature
==== Example ====
I have a table "t" in database "test", we can compare the output with "--flashback" and without.
#client/mysqlbinlog /data/mysqldata_10.0/binlog/mysql-bin.000001 -vv -d test -T t --start-datetime="2013-03-27 14:54:00" > /tmp/1.sql
#client/mysqlbinlog /data/mysqldata_10.0/binlog/mysql-bin.000001 -vv -d test -T t --start-datetime="2013-03-27 14:54:00" -B > /tmp/2.sql
Then, importing the output flashback file (/tmp/2.log), it can flashback your database/table to the special time (--start-datetime).
And if you know the exact postion, "--start-postion" is also works, mysqlbinlog will output the flashback logs that can flashback to "--start-postion" position.
==== Implement ====
1. As we know, if binlog_format is ROW (binlog-row-image=FULL in 10.1 and later), all columns value are store in the row event, so we can get the data before mis-operation.
2. Just do following things:
2.1 Change Event Type, INSERT->DELETE, DELETE->INSERT.
For example:
INSERT INTO t VALUES (...) ---> DELETE FROM t WHERE ...
DELETE FROM t ... ---> INSERT INTO t VALUES (...)
2.2 For Update_Event, swapping the SET part and WHERE part.
For example:
UPDATE t SET cols1 = vals1 WHERE cols2 = vals2
--->
UPDATE t SET cols2 = vals2 WHERE cols1 = vals1
2.3 For Multi-Rows Event, reverse the rows sequence, from the last row to the first row.
For example:
DELETE FROM t WHERE id=1; DELETE FROM t WHERE id=2; ...; DELETE FROM t WHERE id=n;
--->
DELETE FROM t WHERE id=n; ...; DELETE FROM t WHERE id=2; DELETE FROM t WHERE id=1;
2.4 Output those events from the last one to the first one which mis-operation happened.
For example:
- created binlog_encryption test suite and added it to the default list
- moved some tests from rpl, binlog and multisource suites to extra
so that they could be re-used in different suites
- made minor changes in include files
This patch adds DEFAULT as a possible dynamic SQL parameter, e.g.:
EXECUTE IMMEDIATE 'INSERT INTO t1 (column) VALUES(?)' USING DEFAULT;
EXECUTE IMMEDIATE 'UPDATE t1 SET column=?' USING DEFAULT;
and for similar PREPARE..EXECUTE queries.
This is done for symmetry with the STMT_INDICATOR_DEFAULT indicator in
the client-server PS protocol.
The changes include:
- Allowing DEFAULT as a possible option in execute USING clause (sql_yacc.yy)
- Adding "virtual bool Item::save_in_param(THD *thd, Item_param *param)",
because "normal" items (that have real values) and Item_default_value
have now different actions when assigning itself as an Item_param value.
- Fixing switch() statements in a few Item_param methods not to have "default",
because it was easy to forget to add a new "case" when adding a new XXX_VALUE
value into the enum Item_param::enum_item_param_state.
This is important, as we'll be adding new values soon, e.g. for MDEV-11359.
Removing "default" helped to find and report bugs MDEV-11361 and MDEV-11362,
because DECIMAL_VALUE is obviously not properly handled in some cases.
This is similar to MysQL Worklog 3253, but with
a different implementation. The disk format and
SQL syntax is identical with MySQL 5.7.
Fetures supported:
- "Any" ammount of any trigger
- Supports FOLLOWS and PRECEDES to be
able to put triggers in a certain execution order.
Implementation details:
- Class Trigger added to hold information about a trigger.
Before this trigger information was stored in a set of lists in
Table_triggers_list and in Table_triggers_list::bodies
- Each Trigger has a next field that poinst to the next Trigger with the
same action and time.
- When accessing a trigger, we now always access all linked triggers
- The list are now only used to load and save trigger files.
- MySQL trigger test case (trigger_wl3253) added and we execute these
identically.
- Even more gracefully handling of wrong trigger files than before. This
is useful if a trigger file uses functions or syntax not provided by
the server.
- Each trigger now has a "Created" field that shows when the trigger was
created, with 2 decimals.
Other comments:
- Many of the changes in test files was done because of the new "Created"
field in the trigger file. This shows up in SHOW ... TRIGGER and when
using information_schema.trigger.
- Don't check if all memory is released if on uses --gdb; This is needed
to be able to get a list from safemalloc of not freed memory while
debugging.
- Added option to trim_whitespace() to know how many prefix characters
was skipped.
- Changed a few ulonglong sql_mode to sql_mode_t, to find some wrong usage
of sql_mode.
The combination of --remove_file and --write_file on .expect file creates
a race condition which can be hit by MTR which reads the file in a loop.
Instead, .expect file should be changed with --append_file.
It was fixed in 10.x, but in 5.5 the sporadic failure still affected buildbot.
Fixed 3 test files which use the problematic combination
INSERTS/UPDATES ON TEMPORARY TABLES
Bug#14294223: CHANGES NOT ALLOWED TO TEMPORARY TABLES ON
READ-ONLY SERVERS
Problem:
========
Running 5.5.14 in read only we can create temporary tables
but can not insert or update records in the table. When we
try we get Error 1290 : The MySQL server is running with the
--read-only option so it cannot execute this statement.
Analysis:
=========
This bug is very specific to binlog being enabled and
binlog-format being stmt/mixed. Standalone server without
binlog enabled or with row based binlog-mode works fine.
How standalone server and row based replication work:
=====================================================
Standalone server and row based replication mark the
transactions as read_write only when they are modifying
non temporary tables as part of their current transaction.
Because of this when code enters commit phase it checks
if a transaction is read_write or not. If the transaction
is read_write and global read only mode is enabled those
transaction will fail with 'server is read only mode'
error.
In the case of statement based mode at the time of writing
to binary log a binlog handler is created and it is always
marked as read_write. In case of temporary tables even
though the engine did not mark the transaction as read_write
but the new transaction that is started by binlog handler is
considered as read_write.
Hence in this case when code enters commit phase it finds
one handler which has a read_write transaction even when
we are modifying temporary table. This causes the server
to throw an error when global read-only mode is enabled.
Fix:
====
At the time of commit in "ha_commit_trans" if a read_write
transaction is found, we should check if this transaction is
coming from a handler other than binlog_handler. This will
ensure that there is a genuine read_write transaction being
sent by the engine apart from binlog_handler and only then
it should be blocked.
The cause of the issue is when DROP DATABASE takes
metadata lock and is in progress through it's
execution, a concurrently running CREATE FUNCTION checks
for the existence of database which it succeeds and then it
waits on the metadata lock. Once DROP DATABASE writes to
BINLOG and finally releases the metadata lock on schema
object, the CREATE FUNCTION waiting on metadata lock
gets in it's code path and succeeds and writes to binlog.
Creating a CONNECT object on client connect and pass this to the working thread which creates the THD.
Split LOCK_thread_count to different mutexes
Added LOCK_thread_start to syncronize threads
Moved most usage of LOCK_thread_count to dedicated functions
Use next_thread_id() instead of thread_id++
Other things:
- Thread id now starts from 1 instead of 2
- Added cast for thread_id as thread id is now of type my_thread_id
- Made THD->host const (To ensure it's not changed)
- Removed some DBUG_PRINT() about entering/exiting mutex as these was already logged by mutex code
- Fixed that aborted_connects and connection_errors_internal are counted in all cases
- Don't take locks for current_linfo when we set it (not needed as it was 0 before)
Add log_bin_index, log_bin_basename and relay_log_basename system
variables. Also, convert relay_log_index system variable to
NO_CMD_LINE and implement --relay-log-index as a command line
option.
on disconnect THD must clean user_var_events array before
dropping temporary tables. Otherwise when binlogging a DROP,
it'll access user_var_events, but they were allocated
in the already freed memroot.
Update the description of each test case as to why it is
counted in the correct binlog_group_commit_trigger_* variable.
Updated rpl_parallel_multilevel2.test to identify that 3 groups
occur in statement and mixed replication and one group in row based
replication.
remove group_commit_reason_immediate
rename group_commit_reason_transaction to group_commit_trigger_lock_wait
rename group_commit_reason_usec to group_commit_trigger_timeout
rename group_commit_reason_count to group_commit_triggger_count
The following global status variables where added:
* binlog_group_commit_reason_count
* binlog_group_commit_reason_usec
* binlog_group_commit_reason_transaction
* binlog_group_commit_reason_immediate
binlog_group_commit_reason_count corresponds to group commits made by
virtue of the binlog_commit_wait_count variable.
binlog_group_commit_reason_usec corresponds to the binlog_commit_wait_usec
variable.
binlog_group_commit_reason_transaction is a result of ordered
transaction that need to occur in the same order on the slave and can't
be parallelised.
binlog_group_commit_reason_immediate is caused to prevent stalls with
row locks as described in log.cc:binlog_report_wait_for. This immediate
count is also counted a second time in binlog_group_commit_reason_transaction.
Overall binlog_group_commits = binlog_group_commit_reason_count +
binlog_group_commit_reason_usec + binlog_group_commit_reason_transaction
This work was funded thanks to Open Source Developers Club Australia.
Parallel replication (in 10.0 / "conservative" mode) relies on binlog group
commits to group transactions that can be safely run in parallel on the
slave. The --binlog-commit-wait-count and --binlog-commit-wait-usec options
exist to increase the number of commits per group. But in case of conflicts
between transactions, this can cause unnecessary delay and reduced througput,
especially on a slave where commit order is fixed.
This patch adds a heuristics to reduce this problem. When transaction T1 goes
to commit, it will first wait for N transactions to queue up for a group
commit. However, if we detect that another transaction T2 is waiting for a row
lock held by T1, then we will skip the wait and let T1 commit immediately,
releasing locks and let T2 continue.
On a slave, this avoids the unfortunate situation where T1 is waiting for T2
to join the group commit, but T2 is waiting for T1 to release locks, causing
no work to be done for the duration of the --binlog-commit-wait-usec timeout.
(The heuristic seems reasonable on the master as well, so it is enabled for
all transactions, not just replication transactions).
- Fixed compiler warnings
- Added include/wait_for_binlog_checkpoint.inc, as suggested by JonasO
- Updated 'build-tags' to work with git (Patch by Serg)
Using a boolean flag for 'there is a RESET MASTER in progress' doesn't
work very well for multiple concurrent RESET MASTER statements.
Changed to a counter.
