= File-system crash consistency
The guarantee that recovery produces a valid <file system> state under a specified crash model. For the protected <file-system transaction> in our full-data example, the observable <file> is either `ABCD` or `ABCDEFGH`; it never names a free block or exposes a half-applied protected append after recovery. This concerns the recovered state. The intermediate final-location records during a <journal checkpoint> may disagree until <journal replay> finishes, and live concurrent access requires separate <synchronization (computing)>.
Back to article page
