4 ms·
OS crashes do not cause acknowledged writes to be lost. They are already in the drive's queue.
by colanderman 5y ago
OS crashes do not cause acknowledged writes to be lost. They are already in the drive's queue.
- dathinab 5y agoThey do if you don't use F_FULLSYNC, even apple acknowledges it (quote apple man pages): > Specifically, if the drive loses power or the OS crashes, the application may find that only some or none of their data was written. It's also worse then just write losses: > The disk drive may also re-order the data so that later writes may be present, while earlier writes are not.
- colanderman 5y agoI'm using "acknowledged" in the sense of SCSI-level flush (per the thread OP), not mac OS's peculiar implementation of fsync.
- dathinab 5y agoBut the thread OS is about it not being a problem that SCSI-level flushes are supper slow, which is only not a problem if you don't do them (e.g. only use fsync on Mac)? But reading it again there might have been some confusion about what was meant.
- marcan_42 5y agomacOS does issue an NVMe flush on standard kernel panics, although of course it's possible to have it die in a way where that doesn't work.