Hacker News Reader

Comment by cromka | Hacker News Reader

I hope whoever buys their parents, they relicense ZFS

I strongly suspect that broad enthusiasm for copy-on-write filesystems is largely past. Btrfs is out there and is the default for some Fedora editions. While I have no direct connections to that storage world any longer, I'm not convinced that relicensing ZFS would affect the landscape in a major way among the more significant distributions.

It seems that btrfs is being phased out of fedora. It looks like Fedora will be moving to stratis storage over the next year or two. Btrfs is nice but it's raid5/6 support still seems shaky and there's odd corner cases that apply to all btrfs setups (getting stuck when free space is low is mostly resolved but can still occur).

I'm not super-plugged into Fedora these days. Red Hat has long been pretty comfortable with a more conventional file system (think it's still XFS out of SGI by default) plus volume management for RHEL. ZFS was a non-starter for licensing reasons and Oracle; btrfs just wasn't there. I suspect things could have played out differently had Oracle relicensed ZFS once upon a time but that ship has likely sailed in a variety of ways.

XFS is a default filesystem in a production system? Really?

XFS was dead to me 20 years ago. The only time I ever actually verifiably lost data to a system crash was on XFS. I had actual zero length files with XFS after a crash. Lost actual data. But the file system was in totally happy consistent metadata state. Never ever would I ever consider that filesystem ever again.

The ominous "reisserfs" everyone complained about? Never lost a single byte of data to it even when I pulled the plug in the middle of a write operation to test it. No issues with ext4, no experience with btrfs.

The main problem with ReiserFS is Reiser killed his wife so it is unmaintained.

There is that. I think ReiserSF used to be the default that SUSE used. These days, it seems to be btrfs for root and XFS for everything else.

That's why I called it "reisserFS" and not reiserfs. I am well aware of the controversies because of what he did on the non-technical side. I am also of course for a very long time by now not using it any longer.

I was merely talking about the experience with XFS "back in the day" and yes I was on reiserfs and then XFS "came up" and I changed my home partition to that, coz I was looking to move off of reisserfs. Power outage. Boom, lost half my home dir. Suddenly lots of totally fine metadata wise files had null length. I did go back to reiserfs until ext4 was stable enough after that. I will never try XFS ever again, sorry. Loosing data is the thing that a file system can't do to me if they want to "retain my business" so to speak.

And yes, reiserfs I actually tested and had survive some power outages totally fine too. I believed the XFS hype too much without proper testing. My bad I guess in that sense. We were all young once. But also: Never again XFS ;)

You may not believe it but it is true: https://docs.redhat.com/en/documentation/red_hat_enterprise_...

I vaguely know some of the backstory of moving off ext4 (as default). As I understood it at the time, it was less focused on some of the larger systems Red Hat was interested in.

Stratis does not support data integrity checksums. There is dm-integrity hack but it slows down the write speed by halve.

I no longer have the direct relationships with storage folks at Red Hat that I once did and that stuff may all be over on the IBM side anyway. Stratis (which I wasn't familiar with) does seem to be designed with the intention of bringing a lot of pooling etc. on top of a conventional file system but haven't really looked at or talked to anyone in the storage groups recently. Conventional LVM was pretty old and really designed for a different time.

Is it? I thought the contrary, the immutable distros are possible specifically because of them?.

Relicensing it would probably make it into kernel within months and I am absolutely sure adoption would happen quickly. Mind you, ZFS is more important to servers than desktop.

I expect that there is enough history around ZFS that getting it into the kernel would be a pretty big move.

zfs is more mature and has a better feature set. It's also about the best file system for a NAS, since you can detect and repair single bad blocks if you run zfs's version of RAID5/6.

btrfs is lighter on memory use and a bit easier to administer (until it eats your data). But if zfs licensing becomes cleared up and it becomes a normal part of the linux kernel that would be a lot more attractive than btrfs to me

As someone who's used btrfs and zfs, btrfs today is still a pale imitation of what zfs was at launch. btrfs feels like people wanted to make a filesystem which had all the 'pretty neat' features of ZFS but without the tightly coupled management and benefits of ZFS (like volume management + RAID + encryption + compression all being managed through one tool).

btrfs ended up being basically a consumer version of a few of ZFS's most visible features - transparent compression, subvolumes, and copy-on-write - which then farms out the other most useful features - RAID + volume management + encryption - to the existing tooling. Makes sense for a smaller-scale project, no shade, but it definitely misses the mark if you've spent time on ZFS systems taking advantage of what ZFS has to offer (like recursive snapshots or vdevs).

I don't know the whole history of btrfs (although its primary maintainer was I think at Oracle at one point) but ZFS was definitely built from a hierarchical storage enterprise mindset.

Given different historical circumstances, it could probably have picked up more traction. But not sure what would change today even with an Oracle licensing change.

Reply on news.ycombinator.com