bcachefs 1.33.0 - reconcile
20 points by diktomat
20 points by diktomat
Years into development, Kent is still heavily refactoring things under the hood.. and apparently all of this is in preparation for a future rewrite in Rust!
I was very hopeful for bcachefs at one time, but I seriously doubt it will ever be "done" enough to use for anything.
Plenty of people are already using it in what are effectively production scenarios. Criticizing refactoring seems quite frankly bizarre - I'd be much more suspicious of a codebase in real world use that wasn't being periodically reengineered. Similarly rust feels like an obvious direction to move in, and something that couldn't sensibly happen until the rest of kernel started to accept it.
Given how few people are working on the code, and limited funding, the progress rate seems pretty decent. If you want a valid criticism, it might be that low bus factor:(
Weird comment. If anything, ability to do heavy refactoring is a proof of codebase vitality.
I'm using bcachefs for 3? years as a primary fs on my workstation, it saved my data when nvme started dying, and didn't lose any data ever. Much better track record then btrfs I was using for years before then.
Same here, but I've only been using it for about 2 years.
Btrfs has been a disaster the few times I've tried it.
Bcachefs has scared me a few times (error messages from mismatching versions of kernel bcachefs vs userspace bcachfstools) but has never actually lost or corrupted any of my data. Also the few time I've contacted the devs for help (IRC) they've been super nice.
If nothing else, bcachefs has the basics right - a working fsck. btrfs has an fsck that you're basically not supposed to use. I had a drive fail in a data corrupting way this year and whole btrfs filesystems were basically instantly totally unreadable, where as bcachefs (and, tbf, ext4, xfs, etc.) repeatedly managed to recover the vast majority of the data.
I've lost some data to it sadly, after about 2 years of using it. Most of that was on a disk I used to store images, the bcachefs went readonly and corrupted a number of files. I did manage to get most data back, but before all files were rescued, the filesystem became unmountable, requiring me to pull some from backup (sadly the backup hadn't yet captured the newest files). It certainly hasn't been perfect and IMO needs some time to settle and mature, but if you have backups and can rely on them for a restore it's fine.
Years into development, Kent is still heavily refactoring things under the hood.
That shouldn’t be an obstacle for the open source kernel that is infamous for code churn and the inability to keep any internal interface stable.
When I read "bcachefs" and "reconcile", I thought this would have to do with restoring relations with Linus Torvalds and upstreaming the latest bcachefs updates into the mainline kernel.