Generally these are writes that are lost after the kernel confirms data accepted by application write(). The causes can be hardware glitching, the difficulty l

Generally these are writes that are lost after the kernel confirms data accepted by application write(). The causes can be hardware glitching, the difficulty lies with alerting the application that has already moved on assuming all is good. This in the realm of Dan Luu's fsyncgate: https://danluu.com/fsyncgate/ https://danluu.com/fsyncgate/ (discussed here: https://news.ycombinator.com/item?id=19126824 https://news.ycombinator.com/item?id=19126824 ) and "Crash Consistency" https://danluu.com/file-consistency/ https://danluu.com/file-consistency/ The article is about "circling back" and re-checking data blocks against checksums seperately stored rather than relying on hardware checksums and data blocks that match but are "old" (not updated when expected). The "why" of how such things happen is long and turgid exposition into sometimes shit happens .. and that's a book or three by itself.