I’ve come to the conclusion that “write-only” logging filesystems are the way to go. You just write and write and write, with checksums and signatures, but always with a consistent root down view. Then after a crash, find the last “superblock” or whatever that has an intact checksum, Just accept the fact that (the tail end of ) your disk will be inconsistent after a hard boot, and try to find the consistent spot that’s nearby. This is more future proof. Also the way most filesystems are designed these days.
For fsync, you have to splat out a whole pile of data, but it’s linear on disk so it’s fast.
I’ve come to the conclusion that “write-only” logging filesystems are the way to go. You just write and write and write, with checksums and signatures, but always with a consistent root down view. Then after a crash, find the last “superblock” or whatever that has an intact checksum, Just accept the fact that (the tail end of ) your disk will be inconsistent after a hard boot, and try to find the consistent spot that’s nearby. This is more future proof. Also the way most filesystems are designed these days.
For fsync, you have to splat out a whole pile of data, but it’s linear on disk so it’s fast.