COW and volume management? [LWN.net]

COW and volume management?

Posted Jan 28, 2026 10:40 UTC (Wed) by Lennie (subscriber, #49641)
In reply to: COW and volume management? by jmalcolm
Parent article: Filesystem medley: EROFS, NTFS, and XFS

"Btrfs is getting better but not everybody trusts it yet."

Every time I test it, it fails, when I report it on the mailinglist I got no response, I stopped testing it a few years ago.

My test was:

create a VM with 2 disks, set up RAID1 for meta and data, etc.

Scrub and sync them, shutdown the VM, remove disk one and start it again, it usually runs.

Shutdown the VM, connect second disk again.

Start VM and scrub and sync to make sure they are the same again.

Shutdown VM, remove first disk and start VM

Failed. The first time I do this it might work and it would fail on the second time I did the whole cycle again.

A scrub/sync should make sure all the data is on the second disk, why does it fail such a simple test ?

Or is this now fixed ?

to post comments

COW and volume management?

Posted Jan 28, 2026 11:17 UTC (Wed) by koverstreet (subscriber, #4296) [Link] (5 responses)

You might give bcachefs a try. And if you do run into issues, there's a lot of activity on the IRC channel, and we really try to stay on top of bugs and make sure things are working for people.

COW and volume management?

Posted Jan 28, 2026 12:57 UTC (Wed) by koverstreet (subscriber, #4296) [Link] (4 responses)

Also, I should say a bit about methodology when there's issues, for new people jumping into bcachefs.

The expectation isn't that things will be 100% perfect; it's a > 100k loc codebase. You do want your filesystem to be as close to perfect as humanly possible - but developing perfect code in isolation and then releasing it to the world, that isn't possible at the scale we're talking about. There will always be issues, people will always be trying out new and unexpected things, even when it's 100% stable for the situations it was originally developed. We're always pushing boundaries and trying new things, especially as the userbase ramps up.

It's much more important to me that code be debugable than that it be perfect; when something breaks, error messages should be clear enough that even end users can get a decent idea of what went wrong, and the introspection and other debugging facilities need to be complete, thorough and easy to use so that mere mortals can debug in the field.

That's how we make this sustainable, when there aren't enough developers to go around, and that's how we share the load and build up the community and make complex systems understandable and usable.

And we've got a solid community of people who are knowledgable about the system and interested in making sure it works for people - so if you run into issues, you should be able to get answers, especially if you're willing to pay that forward and help the next person out.

COW and volume management?

Posted Jan 28, 2026 23:31 UTC (Wed) by neggles (subscriber, #153254) [Link] (3 responses)

It's really nice to see you sticking to saying positive things about bcachefs, rather than being negative about other filesystems; seems like you're much happier these days :)

COW and volume management?

Posted Jan 29, 2026 1:20 UTC (Thu) by koverstreet (subscriber, #4296) [Link] (2 responses)

Engineering isn't a "good vibes only" club. We have to be able to talk about and understand past mistakes if we want to improve and do better, and sometimes the "what not to do" are really important lessons.

So I can and do talk about that stuff when it's relevant, and no doubt will again.

If you want to avoid the tensions and the drama, the way for that to happen is for everyone to be taking responsibility for their work and responding to the challenges that are arising, because we all depend on each other.

COW and volume management?

Posted Jan 29, 2026 15:30 UTC (Thu) by neggles (subscriber, #153254) [Link] (1 responses)

I completely agree; maybe I should've said "being unnecessarily negative", my bad.

COW and volume management?

Posted Jan 29, 2026 16:08 UTC (Thu) by koverstreet (subscriber, #4296) [Link]

That can be hard one to categorize :) Some people may think everything is fine, and maybe they're right, maybe they're not - and it may not be clear until it's too late.

But I do not think you can look at the kernel community's track record on filesystems and call it a smashing success. There are lessons I think the community has yet to learn. When you see me talking harshly about those things, it's not because I'm trying to be negative or tear the place down - but sometimes people do need tough love.

That thing that makes me happy is just when I see that people are learning, or at least willing to learn. All problems are solvable - in engineering, and programming especially - as long as we're facing up to them.

But yeah, I am happier since the split, and watching the bcachefs community grow and turn into something really special has been a big part of that.

COW and volume management?

Posted Feb 5, 2026 12:55 UTC (Thu) by anton (subscriber, #25547) [Link]

I have done a similar test around 2018, and what we tried, worked (although recovery after a drive failure requires (required?) extra effort). We have been using btrfs in RAID1 configuration since then without problems. I had been using btrfs earlier (starting around 2013 or so) on laptops without RAID1 without problems.