Hacker News Reader

Comment by mustache_kimono | Hacker News Reader

> It's only great due to the lack of competitors in the checksummed-CoW-raid category.

blinks eyes, shakes head

"It's only great because it's the only thing that's figured out how to do a hard thing really well" may be peak FOSS entitlement syndrome.

Meanwhile, btrfs has rapidly gone nowhere, and, if you read the comments to this PR, bcachefs would love to get to simply nowhere/btrfs status, but is still years away.

ZFS fulfills the core requirement of a filesystem, which is to store your data, such that when you read it back you can be assured it was the data you stored. It's amazing we continue to countenance systems that don't do this, simply because not fulfilling this core requirement was once considered acceptable.

View on HN

> Meanwhile, btrfs has rapidly gone nowhere […]

A reminder that it came out in 2009:

* https://en.wikipedia.org/wiki/Btrfs

(ext4 was declared stable in 2008.)

Yes! File systems are hard. My prediction is that it will be *at least* 10 years before this newfangled FS gains both feature- and stability parity with BTRFS and ZFS.

Also, BTRFS (albeit a modified version) has been used successfully in at least one commercial NAS (Synology), for many years. I don't see how that counts as "gone nowhere".

> Also, BTRFS (albeit a modified version) has been used successfully in at least one commercial NAS (Synology), for many years. I don't see how that counts as "gone nowhere".

Excuse me for sounding glib. My point was btrfs isn't considered a serious competitor to ZFS in many of the spaces ZFS operates. Moreover, it's inability to do RAID5/6 after years of effort is just weird now.

raid1c2, raid1c3 and raid1c4 will get you close to RAID5/6 on btrfs (in terms of redundancy), albeit with tad less disk space, but still more than normal raid 1.

> ZFS in many of the spaces ZFS operates

not a lot.

>> My point was btrfs isn't considered a serious competitor to ZFS in many of the spaces ZFS operates.

> raid1c2, raid1c3 and raid1c4 will get you close > not a lot.

I guess I just don't understand this take. btrfs doesn't do what ZFS does, and still isn't as reliable as ZFS is. When it is, maybe I'll take another look. But this is really the problem with btrfs stans -- they've been saying it's ready, when it's not, for years.

Fix the small stuff. Make it reliable. Quit making promises about how it's as good as ZFS, when it's clear it doesn't do all the things ZFS does, just some of the things most of the time.

Years of effort is a stretch, nobody serious (read: who's willing to pay for it) has been working on raid5/6 pretty much since its inception (since nobody serious needs raid 5/6 at all). Western Digital promised to fix it a couple of years ago, but there doesn't seem to be much progress since then.

Are all the foot guns described described in 2021 been fixed?

* https://arstechnica.com/gadgets/2021/09/examining-btrfs-linu...

Not sure about "all", but apart from that article being more pissy than strictly necessary, RAID1 can now, in fact survive losing ore than one disk. That is, provided you use RAID1C3 or C4 (which keeps 3 or 4 copies, rather than the default 2). Also, not really sure how RAID1 not surviving >1 disk failure is a slight against btrfs, I think most filesystems would have issues there...

As for the rest of the article — the tone rubs me the wrong way, and somehow considering a FS shit because you couldn't be bothered to use the correct commands (the scrub vs balance ranty bit) doesn't instill confidence in me that the article is written in good faith.

I believe the writer's biggest hangup/footgunnage with btrfs is still there: it's not zfs. Ymmv.

The author is a Canonical fanboy, a company whom put all their effort into ZFS, which now seems to have been in vain.

> The author is a Canonical fanboy, a company whom put all their effort into ZFS, which now seems to have been in vain.

ZFS is in pretty heavy use?

They put no real effort into ZFS, their own userspace tooling was only half-baked and then thrown aside. Continuing to build and ship the kernel module doesn't cost them much, the hard work of ZFS development is done by others. Quite interesting how you blame others for being fanboys while being a fanboy yourself.

I don't see what's entitled about the idea that "it fulfills the core requirements" is enough to get it "good" status but not "great" status. Even if that's really rare among filesystems.

> I don't see what's entitled about the idea that "it fulfills the core requirements" is enough to get it "good" status but not "great" status. Even if that's really rare among filesystems.

You don't see? Well. Uh, I think this review would make more sense coming from someone who wrote a "great" filesystem, or at the very least understood how hard it was to write ZFS. "Big whoop", or "I don't understand what the big deal is" is what is entitled about it.

If "good" is an accurate assessment, then "It's only great because of lack of competitors" seems like a fair statement to me, and far from "big whoop". The list of problems they put in the post is real and meaningful, and they didn't say it was bad, they implied something more like big fish in a small pond.

> this review would make more sense coming from someone who wrote a "great" filesystem

That's not a reasonable standard for reviewers.

> That's not a reasonable standard for reviewers.

It is when the attitude is "What's the big deal?" ZFS is two decades on, and, is by many metrics, still the state of the art in the traditional filesystem space. What ZFS does is extremely hard, and the reason we know is because every open source competitor can't touch it, so I'm saying -- have a little respect.

You don't like it? You prioritize reflinks (ZFS just merged block cloning BTW, so hello reflinks: https://github.com/openzfs/zfs/pull/13392) or offline dedup over RAIDZ? Fine. But make sure your favorite filesystem (or your new filesystem) can do what ZFS does, day in and day out, before you throw that shade. If it does half the things, or breaks sometimes, it's still a toy compared to ZFS.

I think you read the comment as a lot harsher than it actually was.

Reply on news.ycombinator.com