Recovery

约 392 个字 21 张图片 预计阅读时间 2 分钟

Failure Classification

事务的物理实现

日志记录

事务恢复

一个事务的Commit标志：日志写入稳定存储器

A transaction is said to have committed when its commit log record is output to stable storage

• All previous log records of the transaction must have been output already

这时的修改不一定写到了磁盘

立即修改

The immediate-modification scheme allows updates of an uncommitted transaction to be made to the buffer, or the disk itself, before the transaction commits

redo undo

• redo 历史重演
• undo 需要记录日志 redo-only，完成后，记录\
  • 正常执行过程中的回滚
  • 系统崩溃后恢复发现没有结束

Check Point

fuzzy check point

事务回滚

补偿日志记录

缓冲区管理

log force 将缓冲的日志强行写到Disk

Logical Undo

Redo是按照物理执行

ARIES

• 每个记录一个日志顺序号 LSN

Typically an offset from beginning of log file to allow fast access

• 在数据页中记录一个LSN - Page LSN
  • 更新发生，日志的LSN记录进去
  • 防止 重复的 redo

• Log Record

• Special redo-only log record called compensation log record (CLR) used to log actions taken during recovery that never need to be undone

  • 在回滚时，有助于跳过已经回滚的日志

• Physiological redo

• Dirty Page Table

  • 在后台周期性写回，而不是在Check Point

记录脏页

• Fuzzy Check Point

only records information about dirty pages, and does not require dirty pages to be written out at checkpoint time

流程

• 分析

  • 得到redo的位置RedoLSN —— 至少在Check Point 之前，对脏页Table的Rec取Min
  • undo-list
    从Check Point 往下扫描，完善undo list
  • 更新脏页表

• Redo

  • 脏页表，注意是跟log record中的LSN比较
  • 磁盘页
• Undo