1. consistency
checkpoint region flush 30s or longer, so after crash, the lfs should recovery cr to latest state,eg。read from all inode map, but it doesn't know the latest inode map location,so it need to scan all disk,this incur high overhead。
How does lfs solve this problem?
Lfs record more info,CR know the log end,and one segment can point to next。 so after reboot,it checks whether need to update CR。This tech is know as “roll forward” in db community。
网友评论