3 ms·
Upgrading an ext4 filesystem for the year 2038
- chasil 5y agoI think that conversion to btrfs is another option. The Fedora people seem to like it. https://btrfs.wiki.kernel.org/index.php/Manpage/btrfs-convert https://btrfs.wiki.kernel.org/index.php/Manpage/btrfs-conver...
- quantumofalpha 5y agoZFS is a much better option. btrfs still ate my data as recently as 2-3 years ago. The main thing it has going for it is that it's in the kernel tree. If you don't care about that, ZFS is the right choice today. And if you do care, my money would on bcachefs turning out to be a better in-the-tree contender to ZFS.
- chasil 5y agoZFS is nice, but it won't overlay and upgrade ext4. Fedora decided that improvements since 2-3 years ago are sufficient, and made it the default filesystem (and the world hasn't ended). It's nice to see btrfs mature.
- selfhoster11 5y agoBtrfs ate my data once at a time when ext4 would recover just fine. I'm not giving it another try.
- necheffa 5y agoAs a user of btrfs myself, btrfs is not a great option for single disk workstations. If you look at who is funding development, where they are using btrfs, and the bugs and features getting deferred - its clear the direction of btrfs is not a single disk workstation but a large multi disk system in a rack somewhere.
- chasil 5y agoI'm on the current Oracle Linux 7 with the btrfs layout on the UEKR6, and I've had no problems. I know that it is possible to take a snapshot, yum update, then reboot to the snapshot subvolume to roll back the update, but this seems like a reasonable feature for a single-disk workstation. (I've never tried it, though.) Setting zstd compression per-directory also seems appropriate for single-disk workstations. I can also defrag, which a ZFS single-disk workstation absolutely cannot do.
- KozmoNau7 5y agoI've been using btrfs on both single-disk machines and disk arrays for years. It works well in both setups, the snapshots alone are invaluable. It has withstood flaky disk controllers, dying disks, and power losses, always been able to recover.
- chasil 5y agoThe btrfs filesystem is polarizing, and some have had bad experiences with it, while it has treated others well. Many argue for ext4 or xfs, but so many features are absent from these legacy filesystems. The only place that people appear uniformly happy is FreeBSD.
- billpg 5y agoIs there no way to automatically transition gradually? Is copy/reformat/copy-back really the only way?
- forgotpwd16 5y agoThe inode size is a tunable filesystem parameter and can be changed with tune2fs -I /dev/sda1 As explained in the manual, this rewrites the inode table, is a (or used to be due to inefficient algorithms) lengthy process, and data can be corrupted if it is interrupted. Therefore backups are very much recommended. And if you make a backup (although you should already have) it's simpler and probably faster to just copy back after reformatting.
- pabs3 5y agoThe article uses /boot as an example, but that is usually tiny and requires no backups because it is automatically generated.