Bcachefs Brings Safety Improvements To Linux 6.10, Preps For Online Fsck

[ahrs]

Where does it say in all that licensing gobblygook that YOU can't release a derivative distribution?

I thought that was the intent behind FOSS ... encourage creativity and contribution?

I say Stick It To The Man ... STICK . IT . TO . THE . MAN

If you combine the GPL Linux with ZFS you end up with something that isn't GPL. It's fine if you're distributing external modules but not if it's built-in to the kernel. The two licenses are completely incompatible with each other.

I do it this way because I am not distributing my custom Linux kernel so I don't have to worry about licensing violations. I do it this way on-purpose because I really don't like out-of-tree kernel modules, there's too many ways to shoot yourself in the foot. When I build ZFS in-tree either everything compiles together okay or I'm not upgrading to that kernel.

[PuckPoltergeist]

The controversial slogan has nothing to do with Linus's statement. Stable is different than safe, where safe means won't eat your data. Not stable is reflected in among other things that in Linux 6.9 it already had it's third on-disk format upgrade. Maybe this can help you picture why bcachefs is so safe:

ZFS was promising similar before. And before ZFS it was XFS. And the end was often enough: Recreate your filesystem and restore your backup. So I'm taking my popcorn and will wait

[varikonniemi]

ZFS was promising similar before. And before ZFS it was XFS. And the end was often enough: Recreate your filesystem and restore your backup. So I'm taking my popcorn and will wait

yes, there are many people (me included) who eagerly wait for a report of data loss on bcachefs. If you read my quote you can see how it is highly unlikely. I am no historian, but i don't think any previous general purpose filesystem has been designed to be this resilient.

Got anything remotely similar in design spec from ZFS or XFS? Probably not, only vague marketing speak.

[PuckPoltergeist]

yes, there are many people (me included) who eagerly wait for a report of data loss on bcachefs. If you read my quote you can see how it is highly unlikely. I am no historian, but i don't think any previous general purpose filesystem has been designed to be this resilient.

Got anything remotely similar in design spec from ZFS or XFS? Probably not, only vague marketing speak.

To be honest, XFS never promissed no data loss. It was "only" a mountable filesystem that doesn't need fsck anymore. For ZFS there was the promise of no data loss because of CoW and multi-redundant uberblocks.... For XFS the repair-utility was provided with exhaustive RAM-usage. ZFS has scrubbing and in the end: Restore your backup! Btrfs was the filesystem, where developers were honest: We're aiming for self repair, but we can't provide it for now.
I'm still waiting

[Lexx]

If you mean BCacheFS, I think it'll automatically pick a healthy copy. It just won't go and heal the bad copy(ies) using the healthy copy(ies) / parity. This is all on Kent's radar. I actually asked about including scrub on fsck but I don't recall what he said to that. He is definitely interested in the healing and scrub stuff though.

Not really good that it doesn't repair the second copy (on read) and lacks scrub

[mdedetrich]

Random question, but why is this site so obsessed with bcachefs?

So the current state of new generation CoW filesystems in Linux is not desirable (btrfs has limitations and historical reliability issues for non standard usecases, OpenZFS solves some of these issues but due to its license it cannot sit as an official Linux in tree FS). BCachefs is meant to solve both of these issues so people are understandably eager about it

[woddy]

Could you please elaborate and put it in context? Please...

Stability


Bcachefs describes itself as "working and stable, with a small community of users".^[11] When discussing Linux 6.9-rc3 on April 7, 2024, Linus Torvalds touched on the stability of bcachefs, saying "if you thought bcachefs was stable already, I have a bridge to sell you".^{[12]
https://en.wikipedia.org/wiki/Bcachefs}

[[deXter]]

I've been using BCacheFS since it hit stable in 6.7. Just for Steam stuff, nothing I would be upset at losing.

It has greatly improved since then and the fsck is rock solid. If you use BCacheFS tools 1.7.0 (the latest tag), you can just mount one drive by UUID and it'll use that drive to figure out what others to mount.

I think the overall robustness is solid since 6.9 so far.

The tiering and LZ4 compression is just 🤌
I've put so much crap on the data store and it does really smart tiering. It's as if I have 30 TB of NVME flash storage when the reality is much less than that.

Have you worked with snapshots yet? I wonder if they're any better compared to btrfs, cause right now btrfs is a bit of a PITA when it comes to restoring and deleting root snapshots (like having to manually edit the kernel boot options and fstab to update the subvol ids).

Also wondering how hard it would be to set up a pacman hook to automatically snapshot before an update, and have it appear in my systemd-boot menu.

[blackiwid]

Random question, but why is this site so obsessed with bcachefs?

Well still less focus and hype all BSD people put on ZFS, that was the sole and primary driver for many people using BSD, and it is probably the biggest reason to install some driver that is not included in the kernel under linux.

Compared to that is the focus on such a filesystem under linux minimal.

Still nothing compared to the hype of ZFS:

https://trends.google.com/trends/explore?q=%2Fg%2F11f60xw83w,%2Fm%2F02vw3jq,%2Fg%2F11h23hqcqn

A lot of concentration we had at the merge window because of the drama about it, but since then it's pretty silent about it, maybe slightly more news than about xfs yet xfs is based on a outdated concept COW is where everybody wants to go, so it makes sense that this filesystems get more attention also I would assume because it's less done as btrfs there is more chance happening...

[blackiwid]

Btw has bcachefs has a better solution to not use cow for certain file like virtual machine files or wasn't bigger sqlite also slow in btrfs? Don't remember? Does bcachefs have a automatic solution for this or do the user also manually change stuff?