Migrating away from #bcachefs

https://blog.sesse.net/blog/tech/2025-01-20-21-45_migrating_away_from_bcachefs.html

Steinar H. Gunderson aka Sesse writes:

'"[…] I've converted my last bcachefs filesystem to XFS, and I don't intend to look at it again in the near future. […]

I no longer trust bcachefs' future. […]

[…] I've had catastrophic data loss bugs that went unfixed for weeks despite multiple people reporting them. I've seen strange read performance issues. I've had oopses. […] “oh, yeah, that's a known issue […]

There are more things: […]"'

@kernellogger zfs or nothing. I don't understand people why not to contribute to already existing battle tested solutions, but they rather create their own.

@darukutsu and there are other people that don't understand how Linux users can even think of utilizing or even contributing to a file-system which has some known (albeit often debated) licensing issues and in some people's view interacts badly with core parts of the #Linux kernel…

Side note: The screenshotted comment from https://lobste.rs/s/k6al6k/migrating_away_from_bcachefs#c_ygwogu yet again made me think:

I wonder how may people simply do not known how much XFS has changed in the past decade and that there are quite a few nice XFS features that recently made it into the #Linux #kernel or soon likely will make it – like online repair and a ton of other things, some of which still considered experimental.

But maybe it's also a good thing that this happens without giving XFS new names (like in the EXT[234]).

@kernellogger Not giving it new names is right, it really makes a lot of things hard for the integration with tools (installers, partition mangers and such). There are still the internal format versions which are not compatible.

@kernellogger @kdave If your brand ain't bad, then it's a great strategy. Changing the name is really only useful if you feel you need a new branding effort.

@kernellogger
Or of curiosity - is there any reason to migrate away from ext4 on a normal desktop system?

I installed my Debian ages ago and never thought about filesystems ever since.

@mndflayr if you are happy, unlikely.

But one can get used to subvolumes, snapshots, reflink copies, and other things that Btrfs, XFS and others offer – and once you do it's hard to go back to ext4, unless there are strong reasons to go there.

So if you reinstall or install another machine it likely is worth thinking about I'd say.

@kernellogger

Yeah, it's hard to miss features you don't know about. 🙂

So, for a new volume, rather pick XFS instead of Btrfs?

@mndflayr

Exactly. 😬

XFS or Btrfs: no real opinion there. Maybe someone else here like to share some thoughts.

I'm somewhat familiar with Btrfs, which in the end was the main reason why I avoided CentOS Stream 10 (no Btrfs) on my recently bought home server and chose Fedora Server (XFS default) with Btrfs instead. #YMMV

@kernellogger
I'm a bit surprised that the data loss history of xfs early days stuck harder than, say, ext4 (which def had a rough start...), I guess this is more of a "defaults influence perception" thing in the end

@refi64 yeah, your own perception (reminder, your brain sometimes tricks itself), assumed public opinion, lack of true numbers, and what the media makes you believe is a complicated field hard to navigate. One reason why I'd consider studying psychology is the next life… 🥴

@kernellogger marketing is hard, especially for a filesystem. If I wouldn't follow you and reading some kernel news at phoronix I would know anything about the nice progress in xfs.

@hleithner marketing is hard, especially for open source projects.

A simple page listing newly added experimental, optional, or standard features by kernel and xfsprogs version could do wonders. It's not much work, but some work - and a kind of work that often nobody is willing to do afaics.

Or does something like that exists for XFS?

@mndflayr @kernellogger I went the opposite way, happily used XFS between ~2005 and ~2020, then progressively moved every machine to Ext4 because of fscrypt. The moment it (or something like it) is supported I'll happily go back to XFS or maybe finally try Btrfs...

@_hic_haec_hoc @mndflayr yeah, that might be a reason for ext4.

Hopefully Josef or somebody else picks up fscrypt support for btrfs again sooner or later: https://lore.kernel.org/linux-btrfs/20240411184544.GA1036728@perftesting/

@kernellogger @_hic_haec_hoc @mndflayr Somebody will eventually continue.

@mndflayr @kernellogger i tend to stick with ext4 because after all the years i remember how to turn off sorts of data consistency nonsense and turn it into a real good performer. it also helps that i can understand how to repair it by hand when hardware or brain fails me, which is probably because of good documentation. also, it can be shrunk, which comes handy when reallocating disk space or replacing a failed disk with a slightly smaller one. perhaps these are not very objective reasons, and xfs might actually do better in some of these aspects nowadays, but it's a matter of fact that ext4 is a very reliable, fast, and well understood file system.

@lkundrak @mndflayr not being able to shrink a fs is one or the reasons against XFS for me (it's iirc in the works, but my mind might be tricking me); okay, one can work around it by using dm-thin, but well, that is another layer and thus something I'd like to avoid most of the time

@mndflayr @kernellogger also, i don't think i've ever encountered an ext2/3/4 bug. whereas my first experience with xfs when RHEL switched was that after yum -y update kernel && reboot -f the newly installed kernel was not visible when /boot was on xfs. i suspect this might have been because xfs didn't commit things from the journal journal on unmount to make things faster while grub's read only driver didn't take journal into account. or something like that. it drove me nuts.

one more thing: i find the feature flags in ext4 pretty nice, especially that they can mostly be flipped on and off without reformat. helps with mounting new images on old kernels.

@lkundrak @kernellogger
Thanks. 👍