原文: http://nullwy.me/2018/06/mysq...
若是以爲個人文章對你有用,請隨意讚揚
MySQL 的 binlog 日誌文件,記錄了數據庫表的所有修改操做。本文簡單整理 MySQL binlog 相關知識,以及如何使用 binlog 恢復或閃回數據庫數據。html
要想開啓 binlog,須要在啓動 MySQL 時傳入 --log-bin 參數。或者也能夠在 MySQL 配置文件 /etc/my.cnf
,設置 log_bin
開啓 binlog。MySQL 5.7 開始,開啓 binlog 後,--server-id
參數也必須指定,不然 MySQL 服務器會啓動失敗。java
binlog_format
支持 STATEMENT
, ROW
, MIXED
三種格式,MySQL 5.5 和 5.6 默認爲 STATEMENT
,MySQL 5.7.7 開始默認爲 ROW
。若 SQL 使用 UUID(), RAND(), VERSION() 等函數,或者使用存儲過程、自定義函數,基於 STATEMENT 的主從復時,是不安全的(不少人可能會認爲 NOW(), CURRENT_TIMESTAMP 這些函數也是不安全的,事實上是安全的) [doc1, doc2 ]。基於 ROW
的主從複製,是最安全的複製方式。python
如今先來看下 STATEMENT
格式的 binlog,/etc/my.cnf
文件修改的內容以下:mysql
server_id = 1 log_bin = mysql-bin binlog_format = STATEMENT binlog_row_image=FULL
重啓 MySQL 後,在數據目錄 datadir 下,好比 /var/lib/mysql/
,將會生成相應的 binlog 文件,mysql-bin.index
和 mysql-bin.000001
。.index
後綴的文件保存所有 binlog 文件名。mysql-bin.000001
文件記錄 binlog 內容。每次 MySQL 啓動或者 flush 日誌,都將按序號建立一個新的日誌文件。另外,當日志文件大小超過 max_binlog_size
時,也會建立一個新的日誌文件。git
如今來試一試 binlog 功能。假設在 testdb
庫在有 hello
表,並對其中某行作修改操做:github
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 爲二進制文件,須要使用 mysqlbinlog
(doc, man)命令查看:sql
$ sudo mysqlbinlog /var/lib/mysql/mysql-bin.000001 # 直接在 mysql 服務器上讀取 binlog 文件 $ mysqlbinlog -R -h192.168.2.107 -uroot -p123456 mysql-bin.000001 # 或者,遠程讀取 binlog 文件
執行 update
後相應新增的 binlog 文件內容:shell
# 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/*!*/;
修改 /etc/my.cnf
的 binlog_format
爲 ROW
,再重啓 MySQL。格式修改後,會生成一個新的 binlog 文件 mysql-bin.000002
。數據庫
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,須要使用 sudo mysqlbinlog -v --base64-output=DECODE-ROWS /var/lib/mysql/mysql-bin.000002
命令。執行 update
後相應新增的 binlog 內容:安全
# 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/*!*/;
若執行以下 SQL:
mysql> insert hello (name) values ('Frank'); Query OK, 1 row affected (0.02 sec)
相應生成的 binlog 內容:
# 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/*!*/;
若執行以下 SQL:
mysql> delete from hello where id = 2; Query OK, 1 row affected (0.02 sec)
相應生成的 binlog 內容:
# 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/*!*/;
MySQL 邏輯備份一般會結合全量備份和增量備份,使用 mysqldump
按期全量備份數據庫,而後利用 binlog 保存增量數據。恢復數據時,就是用 mysqldump
備份的數據恢復到備份的時間點。數據庫在備份時間點到當前時間的增量修改,則經過 mysqlbinlog
將 binlog 中的增量數據恢復到數據庫。如今假設已經使用 mysqldump
將數據庫還原到:
mysql> select * from hello; +----+------+ | id | name | +----+------+ | 1 | Andy | | 2 | Lily | | 3 | Will | +----+------+ 3 rows in set (0.00 sec)
以後執行的 SQL:
update hello set name = 'David' where id = 3; insert hello (name) values ('Frank'); delete from hello where id = 2;
無論是使用 STATEMENT
仍是 ROW
,mysqlbinlog
命令均可以將 binlog 增量恢復到數據庫 [doc ]。
觀察 binlog
能夠看到,從最開始的 update hello set name = 'David' where id = 3;
到最終的 delete from hello where id = 2;
,時間上從 "2018-06-17 22:54:13" 到 "2018-06-17 22:56:44",因此基於時間點恢復,命令以下:
$ 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
的事件位置號是從 "154" 到 "956",但須要注意的是 用 --start-position
和 --stop-position
指定位置點範圍,邏輯上對應的是 start <= position < stop
,因此基於時間點恢復,命令以下:
$ sudo mysqlbinlog --start-position=154 --stop-position=957 mysql-bin.000002 | mysql -uroot -p123456
兩種方式任意執行,都能將數據恢復到:
mysql> select * from hello; +----+-------+ | id | name | +----+-------+ | 1 | Andy | | 3 | David | | 4 | Frank | +----+-------+ 3 rows in set (0.00 sec)
binlog2sql,做者爲曹單鋒,大衆點評 DBA。binlog2sql
,從 MySQL binlog 解析出你要的 SQL。根據不一樣選項,你能夠獲得原始 SQL、回滾 SQL、去除主鍵的 INSERT SQL 等。binlog2sql
,底層實現依賴 python-mysql-replication,由該庫完成 MySQL 複製協議和 binlog 格式的解析。
$ 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
生成回滾 sql:
$ 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
閃回的現實原理很簡單,先經過 MySQL 複製協議的 com-binlog-dump 命令 dump 出 binlog,而後按照 binlog 的格式規範解析 binlog,將 binlog 轉換成 SQL,再將這些 SQL 轉換反向邏輯的 SQL,最後再倒序執行。具體能夠看,binlog2sql
做者的文章 [ref ]。
上文中的 binlog2sql
其實底層依賴 python-mysql-replication
庫,這是 Python 庫。若是想使用 Java 解析 binlog 可使用 mysql-binlog-connector-java
(github)庫。目前開源的 CDC 工具,如 Zendesk maxwell、Redhat debezium、LinkedIn Databus 等都底層依賴 mysql-binlog-connector-java
或者其前身 open-replicator。使用 mysql-binlog-connector-java
的示例代碼以下:
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();
輸出(省略部份內容):
... 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] ]}}