MySQL binlog:格式、增量恢复、闪回、Java 解析

MySQL 的 binlog 日志文件,记录了数据库表的全部修改操作。本文简单整理 MySQL binlog 相关知识,以及如何使用 binlog 恢复或闪回数据库数据。

STATEMENT 格式的 binlog

要想开启 binlog,需要在启动 MySQL 时传入 --log-bin 参数。或者也可以在 MySQL 配置文件 /etc/my.cnf,设置 log_bin 开启 binlog。MySQL 5.7 开始,开启 binlog 后,--server-id 参数也必须指定,否则 MySQL 服务器会启动失败。

binlog_format 支持 STATEMENT, ROW, MIXED 三种格式,MySQL 5.5 和 5.6 默认为 STATEMENT,MySQL 5.7.7 开始默认为 ROW。若 SQL 使用 UUID(), RAND(), VERSION() 等函数,或者使用存储过程、自定义函数,基于 STATEMENT 的主从复时,是不安全的(很多人可能会认为 NOW(), CURRENT_TIMESTAMP 这些函数也是不安全的,事实上是安全的)[ doc1, doc2 ]。基于 ROW 的主从复制,是最安全的复制方式。

现在先来看下 STATEMENT 格式的 binlog,/etc/my.cnf 文件修改的内容如下:

1
2
3
4
server_id = 1
log_bin = mysql-bin
binlog_format = STATEMENT
binlog_row_image=FULL

重启 MySQL 后,在数据目录 datadir 下,比如 /var/lib/mysql/,将会生成相应的 binlog 文件,mysql-bin.indexmysql-bin.000001.index 后缀的文件保存全部 binlog 文件名。mysql-bin.000001 文件记录 binlog 内容。每次 MySQL 启动或者 flush 日志,都将按序号创建一个新的日志文件。另外,当日志文件大小超过 max_binlog_size 时,也会创建一个新的日志文件。

现在来试一试 binlog 功能。假设在 testdb 库在有 hello 表,并对其中某行做修改操作:

1
2
3
4
5
6
7
8
9
10
11
12
13
mysql> select * from hello;
+----+-------+
| id | name |
+----+-------+
| 1 | Andy |
| 2 | Bill |
| 3 | Candy |
+----+-------+
4 rows in set (0.00 sec)

mysql> update hello set name = 'Will' where id = 3;
Query OK, 1 row affected (0.02 sec)
Rows matched: 1 Changed: 1 Warnings: 0

binlog 为二进制文件,需要使用 mysqlbinlogdoc, man)命令查看:

1
2
$ sudo mysqlbinlog /var/lib/mysql/mysql-bin.000001  # 直接在 mysql 服务器上读取 binlog 文件
$ mysqlbinlog -R -h192.168.2.107 -uroot -p123456 mysql-bin.000001 # 或者,远程读取 binlog 文件

执行 update 后相应新增的 binlog 文件内容:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
# at 154
#180617 22:47:49 server id 1 end_log_pos 219 CRC32 0x4bd9d69b Anonymous_GTID last_committed=0 sequence_number=1 rbr_only=no
SET @@SESSION.GTID_NEXT= 'ANONYMOUS'/*!*/;
# at 219
#180617 22:47:49 server id 1 end_log_pos 302 CRC32 0x476fafc9 Query thread_id=2 exec_time=0 error_code=0
SET TIMESTAMP=1529246869/*!*/;
SET @@session.pseudo_thread_id=2/*!*/;
SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=0, @@session.unique_checks=1, @@session.autocommit=1/*!*/;
SET @@session.sql_mode=1075838976/*!*/;
SET @@session.auto_increment_increment=1, @@session.auto_increment_offset=1/*!*/;
/*!\C utf8 *//*!*/;
SET @@session.character_set_client=33,@@session.collation_connection=33,@@session.collation_server=33/*!*/;
SET @@session.lc_time_names=0/*!*/;
SET @@session.collation_database=DEFAULT/*!*/;
BEGIN
/*!*/;
# at 302
#180617 22:47:49 server id 1 end_log_pos 423 CRC32 0x7f2c2c7a Query thread_id=2 exec_time=0 error_code=0
use `testdb`/*!*/;
SET TIMESTAMP=1529246869/*!*/;
update hello set name = 'Will' where id = 3
/*!*/;
# at 423
#180617 22:47:49 server id 1 end_log_pos 454 CRC32 0x68da744a Xid = 12
COMMIT/*!*/;

ROW 格式的 binlog

修改 /etc/my.cnfbinlog_formatROW,再重启 MySQL。格式修改后,会生成一个新的 binlog 文件 mysql-bin.000002

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
mysql> show create table hello;
+-------+-------------------------------------------------------------------------+
| Table | Create Table
+-------+-------------------------------------------------------------------------+
| hello | CREATE TABLE `hello` (
`id` bigint(20) NOT NULL AUTO_INCREMENT,
`name` varchar(20) DEFAULT NULL,
PRIMARY KEY (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=4 DEFAULT CHARSET=utf8mb4 |
+-------+-------------------------------------------------------------------------+
1 row in set (0.00 sec)

mysql> select * from hello where id;
+----+------+
| id | name |
+----+------+
| 1 | Andy |
| 2 | Lily |
| 3 | Will |
+----+------+
1 row in set (0.00 sec)

mysql> update hello set name = 'David' where id = 3;
Query OK, 1 row affected (0.02 sec)
Rows matched: 1 Changed: 1 Warnings: 0

查看 ROW 格式的 binlog,需要使用 sudo mysqlbinlog -v --base64-output=DECODE-ROWS /var/lib/mysql/mysql-bin.000002 命令。执行 update 后相应新增的 binlog 内容:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
# at 154
#180617 22:54:13 server id 1 end_log_pos 219 CRC32 0x2ce70d4d Anonymous_GTID last_committed=0 sequence_number=1 rbr_only=yes
/*!50718 SET TRANSACTION ISOLATION LEVEL READ COMMITTED*//*!*/;
SET @@SESSION.GTID_NEXT= 'ANONYMOUS'/*!*/;
# at 219
#180617 22:54:13 server id 1 end_log_pos 293 CRC32 0x8183fddf Query thread_id=2 exec_time=0 error_code=0
SET TIMESTAMP=1529247253/*!*/;
SET @@session.pseudo_thread_id=2/*!*/;
SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=0, @@session.unique_checks=1, @@session.autocommit=1/*!*/;
SET @@session.sql_mode=1075838976/*!*/;
SET @@session.auto_increment_increment=1, @@session.auto_increment_offset=1/*!*/;
/*!\C utf8 *//*!*/;
SET @@session.character_set_client=33,@@session.collation_connection=33,@@session.collation_server=33/*!*/;
SET @@session.lc_time_names=0/*!*/;
SET @@session.collation_database=DEFAULT/*!*/;
BEGIN
/*!*/;
# at 293
#180617 22:54:13 server id 1 end_log_pos 346 CRC32 0x0fc7e1a4 Table_map: `testdb`.`hello` mapped to number 110
# at 346
#180617 22:54:13 server id 1 end_log_pos 411 CRC32 0xb58e729d Update_rows: table id 110 flags: STMT_END_F
### UPDATE `testdb`.`hello`
### WHERE
### @1=3
### @2='Will'
### SET
### @1=3
### @2='David'
# at 411
#180617 22:54:13 server id 1 end_log_pos 442 CRC32 0xef964db8 Xid = 13
COMMIT/*!*/;

若执行如下 SQL:

1
2
mysql> insert hello (name) values ('Frank');
Query OK, 1 row affected (0.02 sec)

相应生成的 binlog 内容:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# at 442
#180617 22:55:47 server id 1 end_log_pos 507 CRC32 0x79de08a7 Anonymous_GTID last_committed=1 sequence_number=2 rbr_only=yes
/*!50718 SET TRANSACTION ISOLATION LEVEL READ COMMITTED*//*!*/;
SET @@SESSION.GTID_NEXT= 'ANONYMOUS'/*!*/;
# at 507
#180617 22:55:47 server id 1 end_log_pos 581 CRC32 0x56f9eb6a Query thread_id=2 exec_time=0 error_code=0
SET TIMESTAMP=1529247347/*!*/;
BEGIN
/*!*/;
# at 581
#180617 22:55:47 server id 1 end_log_pos 634 CRC32 0xedb73620 Table_map: `testdb`.`hello` mapped to number 110
# at 634
#180617 22:55:47 server id 1 end_log_pos 684 CRC32 0x525a6a70 Write_rows: table id 110 flags: STMT_END_F
### INSERT INTO `testdb`.`hello`
### SET
### @1=4
### @2='Frank'
# at 684
#180617 22:55:47 server id 1 end_log_pos 715 CRC32 0x09a0d4de Xid = 14
COMMIT/*!*/;

若执行如下 SQL:

1
2
mysql> delete from hello where id = 2;
Query OK, 1 row affected (0.02 sec)

相应生成的 binlog 内容:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# at 715
#180617 22:56:44 server id 1 end_log_pos 780 CRC32 0x9f52450e Anonymous_GTID last_committed=2 sequence_number=3 rbr_only=yes
/*!50718 SET TRANSACTION ISOLATION LEVEL READ COMMITTED*//*!*/;
SET @@SESSION.GTID_NEXT= 'ANONYMOUS'/*!*/;
# at 780
#180617 22:56:44 server id 1 end_log_pos 854 CRC32 0x0959bc8d Query thread_id=2 exec_time=0 error_code=0
SET TIMESTAMP=1529247404/*!*/;
BEGIN
/*!*/;
# at 854
#180617 22:56:44 server id 1 end_log_pos 907 CRC32 0x2945260f Table_map: `testdb`.`hello` mapped to number 110
# at 907
#180617 22:56:44 server id 1 end_log_pos 956 CRC32 0xc70df255 Delete_rows: table id 110 flags: STMT_END_F
### DELETE FROM `testdb`.`hello`
### WHERE
### @1=2
### @2='Bill'
# at 956
#180617 22:56:44 server id 1 end_log_pos 987 CRC32 0x0c98f18e Xid = 15
COMMIT/*!*/;

使用 binlog 增量恢复

MySQL 逻辑备份通常会结合全量备份增量备份,使用 mysqldump 定期全量备份数据库,然后利用 binlog 保存增量数据。恢复数据时,就是用 mysqldump 备份的数据恢复到备份的时间点。数据库在备份时间点到当前时间的增量修改,则通过 mysqlbinlog 将 binlog 中的增量数据恢复到数据库。现在假设已经使用 mysqldump 将数据库还原到:

1
2
3
4
5
6
7
8
9
mysql> select * from hello;
+----+------+
| id | name |
+----+------+
| 1 | Andy |
| 2 | Lily |
| 3 | Will |
+----+------+
3 rows in set (0.00 sec)

之后执行的 SQL:

1
2
3
update hello set name = 'David' where id = 3;
insert hello (name) values ('Frank');
delete from hello where id = 2;

不管是使用 STATEMENT 还是 ROWmysqlbinlog 命令都可以将 binlog 增量恢复到数据库 [ doc ]。

观察 binlog 可以看到,从最开始的 update hello set name = 'David' where id = 3; 到最终的 delete from hello where id = 2;,时间上从 "2018-06-17 22:54:13" 到 "2018-06-17 22:56:44",所以基于时间点恢复,命令如下:

1
$ sudo mysqlbinlog --start-datetime="2018-06-17 22:54:13" --stop-datetime="2018-06-17 22:56:44" mysql-bin.000002 | mysql -uroot -p123456

binlog 的事件位置号是从 "154" 到 "956",但需要注意的是 用 --start-position--stop-position 指定位置点范围,逻辑上对应的是 start <= position < stop,所以基于时间点恢复,命令如下:

1
$ sudo mysqlbinlog --start-position=154 --stop-position=957 mysql-bin.000002 | mysql -uroot -p123456

两种方式任意执行,都能将数据恢复到:

1
2
3
4
5
6
7
8
9
mysql> select * from hello;
+----+-------+
| id | name |
+----+-------+
| 1 | Andy |
| 3 | David |
| 4 | Frank |
+----+-------+
3 rows in set (0.00 sec)

使用 binlog2sql 闪回

binlog2sql,作者为曹单锋,大众点评 DBA。binlog2sql,从 MySQL binlog 解析出你要的 SQL。根据不同选项,你可以得到原始 SQL、回滚 SQL、去除主键的 INSERT SQL 等。binlog2sql,底层实现依赖 python-mysql-replication,由该库完成 MySQL 复制协议和 binlog 格式的解析。

1
2
3
4
$ python binlog2sql/binlog2sql.py -h192.168.2.107 -uroot -p123456 --start-position=154 --stop-position=957 --start-file='mysql-bin.000002'
UPDATE `testdb`.`hello` SET `id`=3, `name`='David' WHERE `id`=3 AND `name`='Will' LIMIT 1; #start 4 end 411 time 2018-06-17 22:54:13
INSERT INTO `testdb`.`hello`(`id`, `name`) VALUES (4, 'Frank'); #start 442 end 684 time 2018-06-17 22:55:47
DELETE FROM `testdb`.`hello` WHERE `id`=2 AND `name`='Bill' LIMIT 1; #start 715 end 956 time 2018-06-17 22:56:44

生成回滚 sql:

1
2
3
4
$ python binlog2sql/binlog2sql.py --flashback -h192.168.2.107 -uroot -p123456 --start-position=154 --stop-position=956 --start-file='mysql-bin.000002'
INSERT INTO `testdb`.`hello`(`id`, `name`) VALUES (2, 'Bill'); #start 715 end 956 time 2018-06-17 22:56:44
DELETE FROM `testdb`.`hello` WHERE `id`=4 AND `name`='Frank' LIMIT 1; #start 442 end 684 time 2018-06-17 22:55:47
UPDATE `testdb`.`hello` SET `id`=3, `name`='Will' WHERE `id`=3 AND `name`='David' LIMIT 1; #start 154 end 411 time 2018-06-17 22:54:13

闪回的现实原理很简单,先通过 MySQL 复制协议com-binlog-dump 命令 dump 出 binlog,然后按照 binlog 的格式规范解析 binlog,将 binlog 转换成 SQL,再将这些 SQL 转换反向逻辑的 SQL,最后再倒序执行。具体可以看,binlog2sql 作者的文章 [ ref ]。

Java 解析 binlog

上文中的 binlog2sql 其实底层依赖 python-mysql-replication 库,这是 Python 库。如果想使用 Java 解析 binlog 可以使用 mysql-binlog-connector-javagithub)库。目前开源的 CDC 工具,如 Zendesk maxwell、Redhat debezium、LinkedIn Databus 等都底层依赖 mysql-binlog-connector-java 或者其前身 open-replicator。使用 mysql-binlog-connector-java 的示例代码如下:

1
2
3
4
5
6
7
8
BinaryLogClient client = new BinaryLogClient("192.168.2.107", 3306, "root", "123456");
client.setBinlogFilename("mysql-bin.000001");
client.setBinlogPosition(4);
client.setBlocking(false);
client.registerEventListener(event -> {
System.out.println(event);
});
client.connect();

输出(省略部分内容):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
...
Event{header=EventHeaderV4{timestamp=1529247253000, eventType=TABLE_MAP, serverId=1, headerLength=19, dataLength=34, nextPosition=346, flags=0}, data=TableMapEventData{tableId=110, database='testdb', table='hello', columnTypes=8, 15, columnMetadata=0, 40, columnNullability={1}}}
Event{header=EventHeaderV4{timestamp=1529247253000, eventType=EXT_UPDATE_ROWS, serverId=1, headerLength=19, dataLength=46, nextPosition=411, flags=0}, data=UpdateRowsEventData{tableId=110, includedColumnsBeforeUpdate={0, 1}, includedColumns={0, 1}, rows=[
{before=[3, Will], after=[3, David]}
]}}
...
Event{header=EventHeaderV4{timestamp=1529247347000, eventType=TABLE_MAP, serverId=1, headerLength=19, dataLength=34, nextPosition=634, flags=0}, data=TableMapEventData{tableId=110, database='testdb', table='hello', columnTypes=8, 15, columnMetadata=0, 40, columnNullability={1}}}
Event{header=EventHeaderV4{timestamp=1529247347000, eventType=EXT_WRITE_ROWS, serverId=1, headerLength=19, dataLength=31, nextPosition=684, flags=0}, data=WriteRowsEventData{tableId=110, includedColumns={0, 1}, rows=[
[4, Frank]
]}}
...
Event{header=EventHeaderV4{timestamp=1529247404000, eventType=TABLE_MAP, serverId=1, headerLength=19, dataLength=34, nextPosition=907, flags=0}, data=TableMapEventData{tableId=110, database='testdb', table='hello', columnTypes=8, 15, columnMetadata=0, 40, columnNullability={1}}}
Event{header=EventHeaderV4{timestamp=1529247404000, eventType=EXT_DELETE_ROWS, serverId=1, headerLength=19, dataLength=30, nextPosition=956, flags=0}, data=DeleteRowsEventData{tableId=110, includedColumns={0, 1}, rows=[
[2, Bill]
]}}

参考资料

  1. MySQL Replication: 5.1.1 Advantages and Disadvantages of Statement-Based and Row-Based Replication https://dev.mysql.com/doc/mysql-replication-excerpt/5.7/en/replication-sbr-rbr.html
  2. Unsafe Statements for Statement-based Replication https://mariadb.com/kb/en/library/unsafe-statements-for-statement-based-replication/
  3. MySQL 5.7 Reference Manual: 4.6.7 mysqlbinlog https://dev.mysql.com/doc/refman/5.7/en/mysqlbinlog.html
  4. MySQL Internals Manual: 14.9 Replication Protocol https://dev.mysql.com/doc/internals/en/replication-protocol.html
  5. MySQL Internals Manual: 20 The Binary Log https://dev.mysql.com/doc/internals/en/binary-log.html
  6. MySQL闪回原理与实战 https://github.com/danfengcao/binlog2sql/blob/master/example/mysql-flashback-priciple-and-practice.md
  7. https://dev.mysql.com/doc/dev/mysql-server/8.0.11/classbinary__log_1_1Table__map__event.html#details
  8. https://dev.mysql.com/doc/dev/mysql-server/8.0.11/classbinary__log_1_1Rows__event.html#details