满满干货！深入剖析MySQL中可能会忽视的小细节，原来还能这么玩

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#1F1F1F","name":"user"}}],"text":"今日分享开始啦，请大家多多指教~","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"mysql基础","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"对于后端开发来说，打交道最多的应该是数据库了，因为你总得把东西存起来。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"或是mongodb或者redis又或是mysql。然后你发现一个问题，就是他们都有日志系统，那么这些日志用来干什么的呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"举两个例子，回滚和同步。","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"回滚，这里的回滚是比如说一条语句增加了1，然后再减一吗？这里的回滚操作并不是这样的。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"比如说我要更新一条语句，update test set a=1 where b=2,这样的语句，如果这条语句需要回滚，那么操作就应该是在执行前，先查询这条数据进行保存，如果执行完毕需要回滚，那么就直接把原来那条语句写回去。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"又比如说，你的数据库要还原到一个小时前，那么你可以把2个小时前的备份拿出来，然后运行前两个小时到前一个小时的日志文件，那么这个时候就相当于回到了一个小时前了。","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"同步，比如说主从同步了，这样老生常谈的了，一般通过事务日志来同步。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"总之，有了日志，那么可以帮我们实现很多功能的了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么mysql 在innodb 引擎下，有两个日志非常重要，那就是redo log（重做日志）和 binlog（归档日志）日志。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"如果没有这两个日志，应该没啥人敢用mysql的了，因为这两个日志是用了保证mysql的数据完整性的，如果一个数据库连完整性都不能保证，那么是非常危险的。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"redo log","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"首先看下redo log，这个是什么呢？ 这个是innodb存储引擎的日志。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"说一个它的功能哈，前文提及到存储引擎就相当于我们操作系统的文件系统。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么问题来了，我们的文件系统是有缓存的，比方说我们写入一个文件，当我们调用函数的时候，不会直接写入，而是写入缓存去，而后又文件系统自己判断啥时候应该写入进去。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"判断什么时候应该写入进去：","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"其中有一个标准就是这次要写入缓存的时候，判断缓存是否能够装得下，如果装不下，那么先写入文件，清除缓存，然后再写入缓存。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"第二个判断标准就是根据时间某一段时间后进行写入。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"同样存储引擎也要为这一段事情操心啊。如果我们更新一条语句，存储引擎就直接给我们操作正在存储数据的地方，那效率可想而知。所以说，存储引擎就想到一个方法，把更新记录记录到redo log 中，等redo log 快写满，然后就操作到磁盘，或等空闲时间更新进去。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"写完redo log 之后，就会告诉执行器，执行完毕了。这个时候我们的应用程序得到更新成功的回调。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"如果单纯只写入redo log是不行的，因为存储过程不仅要写，还要读啊，如果写完redo log 通知我们的应用程序更新成功，这个时候还没写入到数据文件，那么我们的应用程序去读的时候就读到了旧的数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么这个时候，存储引擎是这么干的。 反正你给我查询的时候要先查询内存，内存中没有才去查询数据文件。那么存储引擎，先更新到内存中，然后更新到redo log。这样对于存储引擎外部来说，是更新了的，毕竟对于外部来说存储引擎是一个整体。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这就是MySQL里经常说到的WAL技术，WAL的全称是Write-Ahead Logging。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么这个redo log 是一个什么样的机制呢？难道就一直记录到一个文件中，然后当要写入数据文件的时候，全部写入，然后删除？如果这么干效率自然是低了。redo log的机制是这样的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"redo log 是由四个文件组成，每个文件大小为1G左右，这个都是可以设置的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"有两个参数，分别为checkpoint 和 write pos。","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"write pos 是当前记录的位置。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"checkpoint 是当前写入到数据文件的位置。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"比如说：","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/11/1107aaa8e8e57ec6691db9a5d8209ac2.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"一开始的时候write pos 写得了第二个文件的问题，如果为位置1000。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这个时候还没有去正在写入数据文件，那么这个时候checkpoint 位置就是0。要往数据文件中写入数据的部分就是checkpoint 到 write pos这一段区间，也就是0-1000位置。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么这个时候存储引擎感觉可以更新了，然后开始写入到正在的数据文件中，那么这个时候开始checkpoints 开始往右移动，假设更新800条。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么就到了下面这个位置：","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/c2/c2971723b3648de6bbe1278625629ee8.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这个时候存储引擎感觉比较忙了，那么就更新800条后，继续接执行器的任务，那么write pos 往右继续移动。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么这个时候就有一个问题出现，比如说文件大小不变，write pos 一直往右移动，这样会超出啊。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么这个时候write pos 发现自己到了末尾，人家又从第一个开始写，覆盖写入。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"如下：","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/58/581f28d128b061b5b3c8fa88253bf36e.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"绿色横线部分是要更新到数据文件的部分。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"redo log 还有一个重要的作用，保证数据正在地写入到数据文件中。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"比如说这个时候正在写入数据文件，然后数据库异常重启了，这里理解异常重启，简单点理解就是内存都没了。那么我们知道写入文件是有缓存的，如果写入到一半异常了，那么数据其实是丢了的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"有了redo log之后，只有数据文件flush了，那么这个时候checkpoint才开始偏移，否则就如果异常了内存没了，那么继续覆盖更新，因为checkpoint 没有变化，那么还是从原来那个异常前的位置开始同步。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么问题来了，这时候就会想，数据文件是文件，redo log也是文件啊。如果写入的时候在缓存区，然后宕机这个时候也没了啊。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"没错，的确有这个问题，这个时候为了数据安全，redo log直接不使用缓存区。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"redo log用于保证crash-safe能力。innodb_flush_log_at_trx_commit这个参数设置成1的时候，表示每次事务的redo log都直接持久化到磁盘。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"binlog","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"binlog 是每个mysql都有的，而不是存储引擎的东西，属于mysql 的server 层的东西。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"对比一下binlog 和 redo log。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这两种日志有以下三点不同。","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"redo log是InnoDB引擎特有的；binlog是MySQL的Server层实现的，所有引擎都可以使用。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"redo log是物理日志，记录的是“在某个数据页上做了什么修改”；binlog是逻辑日志，记录的是这个语句的原始逻辑，比如“给ID=2这一行的c字段加1 ”。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"redo log是循环写的，空间固定会用完；binlog是可以追加写入的。“追加写”是指binlog文件写到一定大小后会切换到下一个，并不会覆盖以前的日志。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"一条更新语句在innodb引擎下的更新过程:","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1. 执行器先找引擎取ID=2这一行。ID是主键，引擎直接用树搜索找到这一行。如果ID=2这一行所在的数据页本来就在内存中，就直接返回给执行器；否则，需要先从磁盘读入内存，然后再返回。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2. 执行器拿到引擎给的行数据，把这个值加上1，比如原来是N，现在就是N+1，得到新的一行数据，再调用引擎接口写入这行新数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3. 引擎将这行新数据更新到内存中，同时将这个更新操作记录到redo log里面，此时redo log处于prepare状态。然后告知执行器执行完成了，随时可以提交事务。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"4. 执行器生成这个操作的binlog，并把binlog写入磁盘。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"5. 执行器调用引擎的提交事务接口，引擎把刚刚写入的redo log改成提交（commit）状态，更新完成。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这个时候有人可能会问，如果第3步骤先更新到内存，这个时候要是读取操作而之后redo log没有写入就宕机怎么办？因为是写入是有锁的，如果没有提交事务，这个时候有写锁。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这时候就发现一个细节，那就是redo log 一条记录有两个状态一个是prepare，一个是commit状态。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么为什么要两个状态呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"我们知道mysql 主从，其实是通过binlog 一条一条发送给从数据库，让从数据库执行binlog里面的操作。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"假设没有这两个状态。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"假如innodb 写入redo log之后呢，这个时候数据库突然宕机了，这个时候redo log 是有的记录的，这个时候binlog 没有记录。那么innodb 通过redo log 进行写入到数据文件后，binlog 依然没有这一条记录。那么从库就少了一条操作了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这个时候主从永远不可能一致。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"如果有了两个状态，数据库重启后，innodb存储引擎还是会通知binlog。这时候两个状态就保证了binlog里面的数据完整性。那么这个时候又会问了，假如上面第四步执行了，第五步没有执行怎么办？比如宕机了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"是啊，这个时候bin log 中有记录但是redo log没有记录。那么从库就少了一条操作记录了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这个时候主从永远不可能一致。同样，我们如果数据库退回到某个时间点，如果binlog 和 redolog不一致的话，同样适用binlog进行回滚一样的会遇到这个问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"如果有了redo log 的prepare 状态，那么如果数据库重启的时候检测到宕机，这个时候redo log里面prepare 状态的数据就会和binlog里面的数据进行校验，进而进行恢复。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这种有两个状态的提交，叫做两阶段提交。他们起到的作用是如果宕机检测到异常，就会对比恢复。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"同样binlog也是文件，同样存在缓存的问题，sync_binlog这个参数设置成1的时候，表示每次事务的binlog都持久化到磁盘。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"mysql 事务","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"事务有四大特性:","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}},{"type":"strong","attrs":{}}],"text":"1.原子性(atomicity)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"一个事务必须被视为一个不可分割的最小单元。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}},{"type":"strong","attrs":{}}],"text":"2.一致性(consistency)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"数据库总是从一个一致性的状态转换到另一个一致性的状态。很多人对事务的一致性和原子性可能会有偏差。要理解这个东西呢，首先要抛开mysql，或者我们常见的数据库sql server，mongodb。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"单纯来理解数据库的事务。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"假如有两个事务，事务a和事务b。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"假设A和B的两个账号，a账户是500块，b账户是300块。里面有一个限制就是A账号不能大于600块。","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"事务a 的逻辑是给A增加100块。然后给B减少100块。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"事务b 的逻辑是给B减少100块。然后给A增加100块。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"现在事务a，开始执行，做好备份做回滚(500,300),然后给A增加了100块，A现在是600。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"现在事务b，开始执行，做好备份做回滚(600,300),然后给B减少了100块，现在B是200。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"现在事务a开始执行，给B减少100，B变成了100。也急速说b事务提交的是(600,100)","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"现在b开始执行，但是报错了，遇到了A不能高于600的限制，现在开始回滚，那么回滚为(600,300)。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么这个时候就变成了(600,300)了。那么请问事务a和事务b 是否符合原子性?","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"首先分析事务a，现在能做的，全部执行了。那么是符合原子性的。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"然后分析事务b，的确是回滚了，也是符合原子性的。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这有疑问吗？没有吧。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么事务a和事务b是否符合一致性呢?","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"a是否符合一致性呢? 数据库总是从一个一致性的状态转换到另一个一致性的状态。一致性的要求是B账户减少100块，A账户多出100块了。这是一致性的要求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"数据库一开始是:(500,300),然后a提交的时候(600,100),这显然不符合。b事务其实是符合一致性的，一开始是(500,300)，回滚也是（500,300），这没错。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么是否一致性就一定要原子性呢?","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/2a/2a11379a14bcba31245d903d50a60f08.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这样其实也是可以一致性的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"那么数据库为什么不这么干呢?比如说 update test set a=a-1;","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这个时候回滚的时候还得给你生成一个update test set a=a+1。这是简单的，如果是复杂的，这是要将数据库称为人工智能吗？数据库都会傻掉的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"要实现一致性，原子性的成本应该是最低的，但是单单原子性是不能实现一致性的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}},{"type":"strong","attrs":{}}],"text":"3.隔离性","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"上面我们知道单单原子性是没有实现一致性的。那么隔离性就是在原子性的基础上增加一些，一些限制条件那么就可以实现一致性。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"比如说，每个事务只能串行执行，这个时候也说符合的，这也是一种隔离级别。但是如果是串行，就不满足并发了，所以就有其他隔离级别了，或者说其他隔离方式。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}},{"type":"strong","attrs":{}}],"text":"4.持久性","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"一旦事务提交，则其所做的修改就会永久保存到数据库中。可能有人说这不是废话吗？事务做的修改不就是要保持带数据文件中，能够持久化吗？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"这里面的持久性，表示得更多的是一种方案。持久化是有很多方式的，怎么确保你的持久化方案可行呢？比如说一个事务要修改4条语句分别在四张表，那么怎么确保这四条语句能够全部写入进去呢？会不会写到第二条的时候系统崩溃呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#333333","name":"user"}}],"text":"如果出现了上面的问题，该怎么处理？这就是持久性的重要性了。前三个都是强调语句执行，最后一个强调存储。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#1F1F1F","name":"user"}}],"text":"今日份分享已结束，请大家多多包涵和指点！","attrs":{}}]}]}