[MySQL 45 lecture -12] lecture 12 the reason why MySQL has a wind attack from time to time

12 | Why my MySQL It's going to shake ？

Preface

• One SQL Sometimes very slow , And the scene is difficult to reproduce , reason ？

• Make a metaphor

  • Powder board ：redo log
  • Shopkeeper's memory ： Memory
  • Account book ： Data files

• Dirty page ： The contents of the memory data page and the disk data page are inconsistent

• Brush dirty pages ： Write memory pages to disk

• Shake it ： Maybe it's really dirty , Usually, a very fast update operation is performed to write memory and logs

What will trigger the database flush The process ？

• 1. The powder board is full of , I can't remember
     （redo log It's full. , The system stops all update operations ）
  • Try to avoid this situation , It's all blocked , The number of updates dropped to 0
• 2. Business is so good , The shopkeeper can't remember
     （ Out of memory , To write dirty pages to disk ）

  • Normal behavior , There should be a mechanism to control the proportion of dirty pages to avoid the following two situations

    • 1. Too many dirty pages to be eliminated , Results in longer response times for queries
    • 2. The log is full of , All updates are blocked
• 3. Update the ledger when business is not busy
     （MySQL When the system is considered idle , Brush dirty pages ）
  • The system is idle , No pressure
• 4. Year end clearing book
     （MySQL Clean the dirty pages during normal shutdown ）
  • The system is idle , No pressure

InnoDB Control strategy of brushing dirty pages

• 1. tell InnoDB Of the host IO Ability

  • innodb_io_capacity

  • Factors affecting the speed of cleaning dirty pages

    • Dirty page ratio
    • redo log Write disk speed
• 2. Continuous sitting mechanism for brushing dirty pages
  • innodb_flush_neighbors
  • When the neighbor of a dirty page is also a dirty page , Will brush together
  • Set the parameter to 0, Turn off the tandem mechanism

Take a chestnut

• One memory configuration is 128GB,innodb_io_capacity=20000 Under the circumstances , It is normal to suggest that you will redo log Set to 4 individual 1GB file . If... Is set during configuration redo log Set to one 100MB file , What will happen ？（ High configuration machine ,redo log Set what happens when it is too small .）

  • Every time a transaction is committed redo log, If the setting is too small, it will soon be filled , The system is forced to stop all updates , What happened ： Disk pressure is low , But the database suffers intermittent performance degradation .