原创转发请注明出处
上接
http://blog.itpub.net/7728585/viewspace-2133188/ 解析MYSQL BINLOG 二进制格式(1)--准备工作
http://blog.itpub.net/7728585/viewspace-2133189/ 解析MYSQL BINLOG 二进制格式(2)--FORMAT_DESCRIPTION_EVENT
http://blog.itpub.net/7728585/viewspace-2133321/ 解析MYSQL BINLOG 二进制格式(3)--QUERY_EVENT
http://blog.itpub.net/7728585/viewspace-2133429/ 解析MYSQL BINLOG 二进制格式(4)--TABLE_MAP_EVENT
http://blog.itpub.net/7728585/viewspace-2133463/ 解析MYSQL BINLOG 二进制格式(5)--WRITE_ROW_EVENT
http://blog.itpub.net/7728585/viewspace-2133469/ 解析MYSQL BINLOG 二进制格式(6)--UPDATE_ROW_EVENT/DELETE_ROW_EVENT
http://blog.itpub.net/7728585/viewspace-2133502/ 解析MYSQL BINLOG 二进制格式(7)--Xid_log_event/XID_EVENT
http://blog.itpub.net/7728585/viewspace-2133506/ 解析MYSQL BINLOG二进制格式(8)--GTID_LOG_EVENT/ANONYMOUS_GTID_LOG_EVENT及其他
http://blog.itpub.net/7728585/viewspace-2133534/ 解析MYSQL BINLOG二进制格式(9)--infobin解析binlog帮助文档
本文解析全部使用工具infobin
参考
http://blog.itpub.net/7728585/viewspace-2133534/ 解析MYSQL BINLOG二进制格式(9)--infobin解析binlog帮助文档
最后我们来回答最开始提出的问题
1、为什么说row格式较statement更占空间
在前面所述 因为row格式记录了真正的数据，比单纯的语句要大得多，
其binlog生成量大约为修改数据量的2/3。如果是update则更大，因为有
前后印象大约为修改数据量的4/3
2、为什么说row格式的binlog更加安全
在前面所述，因为row格式加入了map event，使用table_id进行复制，同时
记录是真正的记录，那么复制是解析的其真正的数据做到了更加安全，
换句话说他是真正的二进制的包含了修改的数据
3、INSERT/UPDATE/DELETE是生成的row binlog如何直接看懂二进制格式
如前面各个文章所描述，并且程序我已经写好了，可以使用参考前面的文章
4、DDL生成的binlog是怎么样的
DDL生成的binlog是语句模式，只有DML操作可以记录行格式。
5、INSERT SELECT/CREATE TABLE 如何生成的row binlog
insert select 生成是binlog是一个事物中的多个event，但是并不是每条数据一个event
而是多条数据生成一个event，一个event大小大约为8K左右。
但是这也有例外就是一行数据本来就超过了8K如下：
insert into testb values(repeat('A',19999));
------>Insert Event:Pos:563(0X233) N_pos:20600(0X5078) Time:1487042557 Event_size:20037(bytes)
create table as 生成的binlog比insert selcet多了一个create table的过程，但是在开启
gtid的情况是不能使用的create table as。
我做了语句：
mysql> insert into testkk select * from test limit 40000;
Query OK, 40000 rows affected (2.52 sec)
Records: 40000 Duplicates: 0 Warnings: 0
如下：
使用
./infobin test.000202|more
>Gtid Event:Pos:356(0X164) N_pos:421(0X1a5) Time:1487040273 Event_size:65(bytes)
Gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1100471
-->Query Event:Pos:421(0X1a5) N_Pos:493(0X1ed) Time:1487040273 Event_size:72(bytes)
Exe_time:0 Use_db:test Statment(35b-trun):BEGIN /*!Trx begin!*/ Gno:1100471
---->Map Event:Pos493(0X1ed) N_pos:666(0X29a) Time:1487040273 Event_size:173(bytes)
TABLE_ID:350 DB_NAME:test TABLE_NAME:testkk Gno:1100471
------>Insert Event:Pos:666(0X29a) N_pos:8822(0X2276) Time:1487040273 Event_size:8156(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:8822(0X2276) N_pos:16982(0X4256) Time:1487040273 Event_size:8160(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:16982(0X4256) N_pos:25143(0X6237) Time:1487040273 Event_size:8161(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:25143(0X6237) N_pos:33304(0X8218) Time:1487040273 Event_size:8161(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:33304(0X8218) N_pos:41455(0Xa1ef) Time:1487040273 Event_size:8151(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:41455(0Xa1ef) N_pos:49572(0Xc1a4) Time:1487040273 Event_size:8117(bytes)
....
------>Insert Event:Pos:5602407(0X557c67) N_pos:5610568(0X559c48) Time:1487040273 Event_size:8161(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:5610568(0X559c48) N_pos:5618719(0X55bc1f) Time:1487040273 Event_size:8151(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
------>Insert Event:Pos:5618719(0X55bc1f) N_pos:5624075(0X55d10b) Time:1487040273 Event_size:5356(bytes)
Dml on table: test.testkk table_id:350 Gno:1100471
>Xid Event:Pos:5624075(0X55d10b) N_Pos:5624106(0X55d12a) Time:1487040273 Event_size:31(bytes)
COMMIT; /*!Trx end*/ Gno:1100471
可以看到大小大约为8K，当然如果不嫌麻烦用mysqlbinlog也行但是event_size需要自己减一下。
6、如果一个delete大表先开始，然后不断有insert小表进入，他是如何生成binlog的。gtid又是按什么顺序生成
我通过分析 delete大表的binlog会在commit后写到binlog中，之前binlog记录在缓存或者临时文件中
但是binlog中的时间记录为事务开始时间为delete发起时间，其gtid生成顺序
是按照commit的时候生成的，同时一个事物的binlog一定是连续的。
如下：
>Gtid Event:Pos:120314(0X1d5fa) N_pos:120379(0X1d63b) Time:1487032811 Event_size:65(bytes) --小事物开始
Gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000922
-->Query Event:Pos:120379(0X1d63b) N_Pos:120451(0X1d683) Time:1487032811 Event_size:72(bytes)
Exe_time:0 Use_db:test Statment(35b-trun):BEGIN /*!Trx begin!*/ Gno:1000922
---->Map Event:Pos120451(0X1d683) N_pos:120503(0X1d6b7) Time:1487032811 Event_size:52(bytes)
TABLE_ID:343 DB_NAME:test TABLE_NAME:testloop1 Gno:1000922
------>Insert Event:Pos:120503(0X1d6b7) N_pos:120543(0X1d6df) Time:1487032811 Event_size:40(bytes)
Dml on table: test.testloop1 table_id:343 Gno:1000922
>Xid Event:Pos:120543(0X1d6df) N_Pos:120574(0X1d6fe) Time:1487032811 Event_size:31(bytes)
COMMIT; /*!Trx end*/ Gno:1000922 --小事物结束
>Gtid Event:Pos:120574(0X1d6fe) N_pos:120639(0X1d73f) Time:1487032783 Event_size:65(bytes) --大事物开始
Gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000923
-->Query Event:Pos:120639(0X1d73f) N_Pos:120711(0X1d787) Time:1487032783 Event_size:72(bytes)
Exe_time:0 Use_db:test Statment(35b-trun):BEGIN /*!Trx begin!*/ Gno:1000923
---->Map Event:Pos120711(0X1d787) N_pos:120888(0X1d838) Time:1487032783 Event_size:177(bytes)
TABLE_ID:342 DB_NAME:test TABLE_NAME:test100009 Gno:1000923
------>Delete Event:Pos:120888(0X1d838) N_pos:129044(0X1f814) Time:1487032783 Event_size:8156(bytes)
Dml on table: test.test100009 table_id:342 Gno:1000923
------>Delete Event:Pos:129044(0X1f814) N_pos:137204(0X217f4) Time:1487032783 Event_size:8160(bytes)
Dml on table: test.test100009 table_id:342 Gno:1000923
------>Delete Event:Pos:137204(0X217f4) N_pos:145365(0X237d5) Time:1487032783 Event_size:8161(bytes)
Dml on table: test.test100009 table_id:342 Gno:1000923
.................
Dml on table: test.test100009 table_id:342 Gno:1000923
------>Delete Event:Pos:102028245(0X614d3d5) N_pos:102036403(0X614f3b3) Time:1487032783 Event_size:8158(bytes)
Dml on table: test.test100009 table_id:342 Gno:1000923
------>Delete Event:Pos:102036403(0X614f3b3) N_pos:102040255(0X61502bf) Time:1487032783 Event_size:3852(bytes)
Dml on table: test.test100009 table_id:342 Gno:1000923
>Xid Event:Pos:102040255(0X61502bf) N_Pos:102040286(0X61502de) Time:1487032783 Event_size:31(bytes)
COMMIT; /*!Trx end*/ Gno:1000923 ---大事物结束
>Gtid Event:Pos:102040286(0X61502de) N_pos:102040351(0X615031f) Time:1487032811 Event_size:65(bytes) --小事物开始
Gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000924
-->Query Event:Pos:102040351(0X615031f) N_Pos:102040423(0X6150367) Time:1487032811 Event_size:72(bytes)
Exe_time:0 Use_db:test Statment(35b-trun):BEGIN /*!Trx begin!*/ Gno:1000924
---->Map Event:Pos102040423(0X6150367) N_pos:102040475(0X615039b) Time:1487032811 Event_size:52(bytes)
TABLE_ID:343 DB_NAME:test TABLE_NAME:testloop1 Gno:1000924
------>Insert Event:Pos:102040475(0X615039b) N_pos:102040515(0X61503c3) Time:1487032811 Event_size:40(bytes)
Dml on table: test.testloop1 table_id:343 Gno:1000924
>Xid Event:Pos:102040515(0X61503c3) N_Pos:102040546(0X61503e2) Time:1487032811 Event_size:31(bytes)
COMMIT; /*!Trx end*/ Gno:1000924 --小事物结束
会话 A 会话B
delete 事物 1487032783开始
+ insert 事物开始 1487032811
+ commit gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000922
+
delete 事物结束commit 生成
commit gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000923
insert 事物开始 1487032811
commit gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000924
实际上记录binlog的顺序为
insert commit gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000922
delete commit gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000923
insert commit gtid:4a6f2a67-5d87-11e6-a6bd-0c29a879a3:1000924
也就是说delete记录到了第一个insert之后，按照是commit的顺序来记录和分配gtid的
也就是上面解析出来的结果。
至此mysql binlog 二进制格式解析系列文章告于段落
