[ThatPlayer]

The article mentions bcache, but not bcachefs which you can also do this on. I also like bcachefs more for mixed drive type filesystems because of the bcache feature of using the SSDs as write caches for data that gets moved to the HDD in the background. Also read cache.

This btrfs feature only supports having the data chunks prioritize the HDDs.

Whether or not that matters depends on your use case of course: it's not worse than a RAID setup of only HDD. I'm using bcachefs on budget gaming PCs built from old spare parts, so having a small SSD and a bunch of HDDs for games is great.

[raj_hun]

I like the idea behind bcache, sadly it was never mainlined in the upstream kernel, because the maintainer is generally uncooperative with the community and is building his own walled garden because btrfs was "Not Invented Here".

[homebrewer]

I'm pretty sure Kent explained how btrfs's problems lie in its design and are fundamentally unfixable. It's got nothing to do with NIH.

[kjs3]

No. Kent explained in his transparently biased opinion why btrfs is fundamentally unfixable. Which include such gems as "they wrote it too fast to get it right" (opinion much?) and "it has more code than XFS" (FS that tries to do more has more code is bad?). Many people, including people who just might know a few things about filesystems, have addressed his claims, both in detail and in the large. Now, I am in absolutely no position to objectively evaluate the detailed claims of dueling filesystem gurus, but having read quite a bit, I do feel pretty confident in thinking "it's complicated" and "btrfs for all its issues certainly isn't the flaming shitshow Kent really, really wants everyone to believe it is".

[koverstreet]

If an opportunity arises to respond to this elsewhere, I will. I don't think I'll dump on btrfs in a btrfs thread, though :)

Keep in mind though I wrote all that ~10 years ago, and it was very much an opinionated mission statement. bcachefs started out 15 years ago as a couple of us sitting around in the office drinking beers "you know, this looks like a really elegant basis for a filesystem, it's going to be way smaller and cleaner!" (talking utter shit, of course). But, somehow, I stuck with it...

[kjs3]

That's nice. The point being a 10 year old, self-confessed 'opinionate mission statement' which has in the intervening years seen some notable pushback (and perhaps progress in the opposition) should not be trotted out as the grandparent did and presented as a case of cadit quaestio.

[koverstreet]

Because?

I don't see anything there that's fundamentally changed. Btrfs repair code is still quite weak, and "btrfs ate my data" stories have reduced somewhat, but I'm still seeing them come up, and the core architectural issues haven't changed. Josef was recently saying on lkml that they (the btrfs team, at least at facebook) have accepted that XFS will always be better in many scenarios, and to me that's an admission of failure - a COW filesystem should work just as well as a previous generation non-COW filesystem in nocow mode.

A mission statement like that is fundamentally saying "these are my priorities, this is why we need to do better" and I stand by what I wrote, because taking on writing a new filesystem - the part of the operating system with the highest requirements for "this absolutely has to work" robustness - really is about having the right priorities.

Gotta take your time your time to get things right, make sure the fundamentals are solid, before chasing all the features - this isn't an area where typical silicon valley "go big or go home, fake it till you make it" attitudes can work.

[koverstreet]

bcache and bcachefs are both mainlined - please don't spread this sort of FUD.

I'm taking the experimental label off in 6.16, so it'd be nice to avoid the drama and have a smooth release :)

(also, fun to see btrfs copying bcachefs)