Ubuntu's ZFS Installation Work Will Continue Into The 20.04 LTS Cycle

[starshipeleven]

Aint there licensing problems with CDDL and GPL if you ship kernel modules, does the installer compile the kernel modules on request?

No, CDDL only prevents to integrate the module in the kernel source, aka upstreaming. Shipping the module in a distro is OK.

[alcalde]

Yeah sure, the ability to seamlessly rollback any update or system change, make backups and such with an online system, be able to compress the filesystem contents (which can decrease disk space usage significantly, and if you use VMs it's even more important) is something no consumer wants.
...Also totally not what OpenSUSE is already doing with btrfs. Nothing to see there, move along. ZFS is only for enterprise.

Well that's the point... you can already do all that with btrfs, so what's the point of putting all the work into doing it all over again with ZFS? Canonical has never learned to work with others.

[DanL]

Well that's the point... you can already do all that with btrfs, so what's the point of putting all the work into doing it all over again with ZFS?

Because ZFS predates btrfs and has a better track record.

Canonical has never learned to work with others.

How is enabling support for ZoL (in addition to all the other choices, including btrfs) in the installer an example of not working well with others? Either find facts that back your assertion or take your FUD elsewhere.

[starshipeleven]

Well that's the point... you can already do all that with btrfs, so what's the point of putting all the work into doing it all over again with ZFS?

They want to have their own "signature" product. And to be fair ZFS does have a better track record so far. Btrfs still can't do RAID5/6 reliably and this pisses off people to no end.

And before we all flame Canonical for going their own way, I'd like to point out that they are probably the more conservative ones in the bunch, as ZFS is the more proven here.

RedHat went the "full retard" way and decided they would create something new and better with XFS extensions, LVM and dm-crypto, and call this Stratis. Could be amazing or not. I personally really hope it turns well.
OpenSUSE is betting on Btrfs.
Oracle is offering their respin of CentOS with Btrfs.

[alcalde]

How is enabling support for ZoL (in addition to all the other choices, including btrfs) in the installer an example of not working well with others? Either find facts that back your assertion or take your FUD elsewhere.

Well, doing something alone isn't an example of working well with others, is it? If they want these things, let them contribute to btrfs, work with OpenSUSE, etc. Heh, I remember when OpenSUSE tried to offer Unity and eventually gave up, publicly posting that there were so many forks of so many libraries, so little of which was documented, that they found it impossible to get running.
This is another example of the entire Linux community going one way and Canonical seemingly deciding to duplicate the work of others rather than collaborate. As for the accusation of FUD, I've written a novella-length catalog of Canonical's lack of cooperation in the past. In fact, at this point, to suggest that Canonical has a history of cooperation and collaboration with others in the Linux community would be a Sagan-esque extraordinary claim that requires extraordinary evidence.

[DanL]

Well, doing something alone isn't an example of working well with others, is it?

No, but it doesn't mean you refuse to or aren't capable of working with others...

If they want these things, let them contribute to btrfs, work with OpenSUSE, etc.

Why would they do that if their users/customers prefer ZFS? That would be wasted/duplicated effort.

This is another example of the entire Linux community going one way and Canonical seemingly deciding to duplicate the work of others rather than collaborate.

Sigh. They're not duplicating effort. They're just putting support for ZFS in their installer. Lots of distros have unofficial packages of ZoL, and they could follow Ubuntu's example.

As for the accusation of FUD, I've written a novella-length catalog of Canonical's lack of cooperation in the past. In fact, at this point, to suggest that Canonical has a history of cooperation and collaboration with others in the Linux community would be a Sagan-esque extraordinary claim that requires extraordinary evidence.

I'm not claiming there aren't instances where they made some headscratching moves that seemed like creation for the sake of NIH. But some of their innovations were useful, and there are also plenty of instances of Canonical collaborating with others and contributing upstream. In fact, I'd say they've gotten a lot better at it lately (contributing to gnome-shell, helping MATE with Wayland support, etc.)
To simply make the statement that "Canonical never learned to play with others" is pure BS. I wouldn't even call it a half truth.

[rleigh]

RedHat went the "full retard" way and decided they would create something new and better with XFS extensions, LVM and dm-crypto, and call this Stratis. Could be amazing or not. I personally really hope it turns well.

Hopefully it will work well. But I'm very sceptical. RedHat effectively own LVM (via Sistina) and XFS. They clearly want to maximise their investments in the companies and technologies involved, since they employ a number of very experienced developers for all these parts. However... LVM is very dated technology. XFS is a decade older again. No matter how well you cobble these bits together with glue and duct tape (Python and D-BUS), LVM with thinpools and XFS with a bunch of extensions to allow it to communicate with the other layers, is never going to surpass ZFS in terms of features, performance or reliability. I took a detailed look at it a few months back, and it's a poor man's ZFS with a tiny fraction of its capabilities.

I do wish them all the best with it. But it's going to be an uphill battle for them to push this on a userbase who can see full well what its limitations are compared with the competition. I do wonder exactly who they are going to pitch this technology to. Is it solely RHEL customers who are locked into RedHat technologies irrespective of their actual merit?

[skeevy420]

Hopefully it will work well. But I'm very sceptical. RedHat effectively own LVM (via Sistina) and XFS. They clearly want to maximise their investments in the companies and technologies involved, since they employ a number of very experienced developers for all these parts. However... LVM is very dated technology. XFS is a decade older again. No matter how well you cobble these bits together with glue and duct tape (Python and D-BUS), LVM with thinpools and XFS with a bunch of extensions to allow it to communicate with the other layers, is never going to surpass ZFS in terms of features, performance or reliability. I took a detailed look at it a few months back, and it's a poor man's ZFS with a tiny fraction of its capabilities.

I do wish them all the best with it. But it's going to be an uphill battle for them to push this on a userbase who can see full well what its limitations are compared with the competition. I do wonder exactly who they are going to pitch this technology to. Is it solely RHEL customers who are locked into RedHat technologies irrespective of their actual merit?

Do you have any idea how "dated" ZFS is? There are a lot of arguments to pick on these positions. That's not a very good one in this instance.

It's for people who want and like ZFS but don't like having to deal with all the ZFS hoops and hurdles since it isn't treated as a first class citizen in LinuxLand.

[starshipeleven]

Do you have any idea how "dated" ZFS is? There are a lot of arguments to pick on these positions. That's not a very good one in this instance.

It's for people who want and like ZFS but don't like having to deal with all the ZFS hoops and hurdles since it isn't treated as a first class citizen in LinuxLand.

Quite frankly, most of ZFS "hoops and hurdles" are actually in using it, due to ZFS's own architecture. It's not hard to ship a known-working default system with ZFS if you are a distro maintainer.

[skeevy420]

Quite frankly, most of ZFS "hoops and hurdles" are actually in using it, due to ZFS's own architecture. It's not hard to ship a known-working default system with ZFS if you are a distro maintainer.

I just meant the bootloader features crap, having to be particular about kernel updates in case the module doesn't compile, no support with Dolphin/KDE and other little things like that, dealing with IO schedulers and udev rules, possibly having to compile and maintain it yourself since very few distributions provide it, etc.

For an end-user wanting to use ZFS for more than a non-root based data storage solution, there's a lot of studying up that needs to be done so you don't hit a FUBAR scenario and have to start over*. One wrong zpool upgrade and the bootloader is hosed. Most of hoops and hurdles mainly effect us legacy people stuck with GRUB -- having to deal with both that and actual ZFS stuff gets old and my frustration shows it

*Granted that that's true to take full advantage of any advanced file system and not limited to ZFS.