Linus Torvalds Doesn't Recommend Using ZFS On Linux

[wikinevick]

Please read this post to understand CDDL was intentionally based from MPL 1.1. known for its incompatibility to GPL in addition of removing some lines to prevent porting tools to Linux kernel. The main topic is about Linus refusing to support codes from openZFS community inside the kernel space as long Oracle lawyers and CEO will not clarify the legal issue related ZFS. Until then, openZFS community are on their own fixing when Linux kernel broke their codes in user space .

BS, the GPL incompatibililty is a mis-interpretation from 1 ex-SUN lawyer. Ask Brian Cantrill or the people involved in OpenZFS, and they have no issues with OpenZFS being used in linux. Oh .. and Canonical has lawyers too.

[nivedita]

Btrfs and Zfs are unstable and Zfs is unsupported as well. That's the problem. In order to detect bit rot with Linux you have to take enormous risks with the data itself because of unstable and unsupported file systems. That's why I said it was a shame.

ZFS is both stable and supported. It's just not supported by the Linux kernel developers, but by a different set of people. Who happen to work at Lawrence Livermore Labs.

[muncrief]

ZFS is both stable and supported. It's just not supported by the Linux kernel developers, but by a different set of people. Who happen to work at Lawrence Livermore Labs.

Yes, I saw an earlier link to something about that. The problem is that I constantly read articles like this one about it breaking, and I don't have an IT department like Lawrence Livermore Labs to deal with it.

I've also tried setting up single disk ZFS systems a few times before and could never get it to work in any case. For example, last time I simply wanted to format 3 individual disks with ZFS, but no matter what I did I suffered data corruption as soon as I started using them. I'm an embedded systems designer and know quite a bit about hardware, software, and firmware but I simply couldn't figure out what was wrong. It seems ZFS is designed to be used in highly complex systems, and if you just want to use it on individual disks it doesn't work.

By the way, I use mergerfs for disk pooling because it's far more versatile than anything else I've ever tried. So as I said in my OP all I need is a file system capable of detecting bit rot. But after all these years I've still never found one.

[nivedita]

Yes, I saw an earlier link to something about that. The problem is that I constantly read articles like this one about it breaking, and I don't have an IT department like Lawrence Livermore Labs to deal with it.

I've also tried setting up single disk ZFS systems a few times before and could never get it to work in any case. For example, last time I simply wanted to format 3 individual disks with ZFS, but no matter what I did I suffered data corruption as soon as I started using them. I'm an embedded systems designer and know quite a bit about hardware, software, and firmware but I simply couldn't figure out what was wrong. It seems ZFS is designed to be used in highly complex systems, and if you just want to use it on individual disks it doesn't work.

By the way, I use mergerfs for disk pooling because it's far more versatile than anything else I've ever tried. So as I said in my OP all I need is a file system capable of detecting bit rot. But after all these years I've still never found one.

The article is a bit misleading. Nothing was actually ever broken with zfs on linux. What happened was that the latest version of zfs would not compile on the latest version of the kernel for a small period of time.

I don't know what went wrong for you, but I've run root on ZFS with ZFS from the git tip, and kernel either latest stable or git tip, and have never suffered data corruption -- there have been crashes, but never data loss. ZFS on a single disk should work just fine as long as something else isn't messing with that disk -- for eg, once I had an issue where I was installing Windows for dual-boot, and it went and decided to install itself on the disks with the zfs pool instead of the boot disk I wanted it to use. But a reboot into linux followed by zfs resilver fixed that with no data corruption.

[itoffshore]

I've also tried setting up single disk ZFS systems a few times before and could never get it to work in any case. For example, last time I simply wanted to format 3 individual disks with ZFS, but no matter what I did I suffered data corruption as soon as I started using them. I'm an embedded systems designer and know quite a bit about hardware, software, and firmware but I simply couldn't figure out what was wrong. It seems ZFS is designed to be used in highly complex systems, and if you just want to use it on individual disks it doesn't work.

When you create a ZFS pool with a single disk you should set option "copies=2" so data can be kept consistent (also useful for ZFS on external USB drives)

[madscientist159]

Uh, 4.14 about. https://btrfs.wiki.kernel.org/index.php/Gotchas

We use BTRFS on a number of production systems. Unfortunately, we lost a small amount of data on a key system running 5.2 from this bug (window from last backup to system crash):

https://bugs.archlinux.org/task/63733

We're still using it because we need the advanced features, but confidence is definitely shaken after that issue. Storage subsystem was fine, but BTRFS just wasn't writing the data out correctly.

[muncrief]

When you create a ZFS pool with a single disk you should set option "copies=2" so data can be kept consistent (also useful for ZFS on external USB drives)

But then you lose half the capacity of the drive. As I said in my OP, I simply want a filesystem that can detect bit rot. I don't want to waste disk space duplicating data, because that's what my cloud backups are for. And I don't want to have to do any other ridiculous things like having identical disks, rebuilding everything when I want to add/delete storage, etc.

I simply want to format my individual disks, use mergerfs to pool them in a flexible and simple manner, and detect bit rot and restore files from backups when it occurs.

It's really quite simple, but you can't do it with Linux. And as I've said a few times now, it really is a shame.

[finalzone]

BS, the GPL incompatibililty is a mis-interpretation from 1 ex-SUN lawyer. Ask Brian Cantrill or the people involved in OpenZFS, and they have no issues with OpenZFS being used in linux. Oh .. and Canonical has lawyers too.

As highlighted on comment #52, Brian Cantrill was not at the room when the decision to make CDDL intentionally incompatible to GPL was one of the main reasons to prevent porting the codes to Linux kernel space. Only Oracle know what OpenZFS could infringe because they own ZFS and wait for the opportunity to strike as the lawsuit against Google related to Java showed. Should you carefully ready the post, the legal issue is related to kernel space from which Linus refuses to support OpenZFS codes.

Canonical can only use OpenZFS on user space (possibly advised by their lawyers) and they are on their own fixing the problem when an update from kernel space breaks those codes which were never supported.

[wikinevick]

You don't know nothin' John Snow,. Really .. I won't even spend my time explaining.

[finalzone]

You don't know nothin' John Snow,.

Neither do you.

Really .. I won't even spend my time explaining.

You already spent time posting claims getting refuted and are afraid to concede to save face. After all, we are still learning every day.