本文详细介绍了数据库系统中内存页修改和磁盘同步的过程。首先，修改发生在内存中的页面，被标记为脏页。接着，修改的操作会在本地微型事务缓冲区生成重做日志。随后，重做日志记录会被写入全局内存的重做日志缓冲区，并最终写入磁盘。确保重做日志先于脏页刷新到磁盘，遵循写前日志（WAL）原则，以保证数据一致性。脏页会在后续的检查点操作中被刷新到磁盘。
Any changes to a page is first done to the in-memory copy of the page. The page that is modified in memory and not yet flushed to disk is marked as the dirty page in memory.
An associated redo log is generated in memory, in the local mini transaction (mtr) buffer. This will then be transferred to the global in-memory redo log buffer.
The redo log record is written from the redo log buffer in memory to the redo log file on disk. This is subsequently flushed. These two steps are considered separate - writing to the redo log file and flushing the redo log file to the disk. This is to account for the file buffering done by the operating system.
The dirty page is flushed from memory to disk at some later point of time as part of the checkpointing operation.
The order of these steps are important. The redo log record of a change must be flushed to disk before flushing the corresponding dirty page to the disk. This is the concept of write-ahead logging (WAL).
扫一扫
108
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
