Some of this is a repeat, as this comes up from time to time. Neither Red Hat nor Fedora currently have Btrfs developers to manage any Btrfs regressions that come up. And as the upstream development is still very busy, there are regressions, and there's every reason to believe Fedora users would get disproportionately impacted because Fedora kernels tend to be very recent.
SUSE is in a sweet spot because they have a bunch of Btrfs developers, and they are running longterm kernels. So it is upstream testers, including Fedora using testers, who hit regressions, get them reported and fixed before they could ever appear in a longterm kernel on any of the SUSE offerings. And because of their strong Btrfs developer presence, they're doing a lot of stable backports. I think Fedora could soon do Btrfs with single device only. But I think it requires community work to build Btrfs regression tests (these exists already) within Fedora infrastructure, so that every built kernel runs those tests. And then some interpretation is needed to know whether a particular failure is tolerable, users need notification, or the build needs to be failed. The details of this need to be reviewed by the kernel team. Right now they can't do that work themselves, they have plenty on their plate now. As for stability, it's kinda complicated. Single device Btrfs is stable, unless the device lies about committing to disk when it hasn't, and as it turns out devices do transiently corrupt data. Btrfs is going to give a heads up when that happens, and if it's bad, as in further writes can corrupt the filesystem, it'll go read only. Other filesystems tolerate this condition far longer. What's interesting is both XFS and ext4 now default at mkfs time to checksumming metadata. So they can catch similar problems, just not corruption with actual file data which is a much larger target, and can still cause further system corruption if it goes unchecked. Anyway, since using it as primary filesystem for roughly 5 years, I've had no problems on single device Btrfs that weren't user induced. On raid1 and raid 10 I have found meaningful bugs and deficient features, that a user can innocently run into, despitere being known and documented. But in those cases, while i lost redundancy, no data was lost or corrupted. Meanwhile the Btrfs list still fields weird failure in occasion. Steps for repairing Btrfs if it face plants is really non obvious. It's a lot like throwing spaghetti at a wall (even though that's not how you should test your spaghetti people!) Eventually the idea is, kernel code should not f up in the first place even if the device lies, but even if it goes badly, can recover and fix itself without fsck. The fsck is really something of a debugging and time saving tool. Not the ideal scenario. Consider that Btrfs is supposed to scale, and none of the current raid levels scale, and fsck doesn't scale either, nor does scrub, or balance. So Btrfs is going to keep changing and getting better. -- Chris Murphy
_______________________________________________ users mailing list -- users@lists.fedoraproject.org To unsubscribe send an email to users-le...@lists.fedoraproject.org
