ZFS On Linux 0.8.3 Released With Many Fixes

[starshipeleven]

Strangely, I don't see the post you are replying to so I reply to you. I was a huge fan of btrfs. Then I found ZFS and never looked back. ZFS is not perfect and yes, btrfs has some minor advantages over ZFS. But overall, ZFS is the best filesystem for people who value their data.

If you want to use raid5/6 (aka raidz-something in zfs), yes it is better.

I personally like a lot to be able to make a RAID1 with random assorted drives of different sizes I have in btrfs, as it allows me to recycle old stuff.

[Britoid]

I can't wait for it to break next week.

[lu_tze]

Of course, for any specific need you might find a specific filesystem that is a better fit. But if you don't appreciate ZFS you have never used it extensively. You have to spend quite a lot of time to get to know its quirks. And the information is not always readily available from one source (the best source seem to be a couple of books by Michael Lucas and Allan Jude - I strongly recommend them for anyone serious about ZFS). But at the end of the day, there just isn't anything that is remotely close to its quality and features.

BTW, IMHO, Linus was talking out of his ass on the technical aspects.

Well, if you used it extensively, you would know, that it's quirks are also something basic like not being able to mount volumes at startup (see: https://github.com/zfsonlinux/zfs/issues/8885). It is kind of bummer, if your installation has zfs root and it cannot find it's own root volume.

[skeevy420]

Well, if you used it extensively, you would know, that it's quirks are also something basic like not being able to mount volumes at startup (see: https://github.com/zfsonlinux/zfs/issues/8885). It is kind of bummer, if your installation has zfs root and it cannot find it's own root volume.

Not knowing shit like enabling the systemd units is no different than not knowing about LUKS2 headers and GRUB or BTRFS and swapfiles when using older kernels.

Personally, I consider what you linked to to be user error from a lack of knowledge and not a problem with ZFS.

[lu_tze]

Not knowing shit like enabling the systemd units is no different than not knowing about LUKS2 headers and GRUB or BTRFS and swapfiles when using older kernels.

Personally, I consider what you linked to to be user error from a lack of knowledge and not a problem with ZFS.

What you consider it is of no consequence; the maintainers don't, otherwise they would be able to close the issue.

It also has nothing to do with systemd units (despite some trying to flip any tunable they can), because the problem happens much sooner - it cannot find the rootfs, got it? That happens way before systemd looks up which units to launch. Oh, and it is broken with 0.8.x only, 0.7.x works. It also means that there are machines that cannot update, which never happen with any other filesystem, not even btrfs.

[skeevy420]

What you consider it is of no consequence; the maintainers don't, otherwise they would be able to close the issue.

It also has nothing to do with systemd units (despite some trying to flip any tunable they can), because the problem happens much sooner - it cannot find the rootfs, got it? That happens way before systemd looks up which units to launch. Oh, and it is broken with 0.8.x only, 0.7.x works. It also means that there are machines that cannot update, which never happen with any other filesystem, not even btrfs.

Don't add "not even BTRFS" to that. Don't update your kernel because a new BTRFS bug will destroy your data has happened before and will likely happen again.

Based on reading that page again, it is still mostly user error, distribution startup differences, lack of user knowledge, some outdated documentation, and, in the case of the OG poster, was because they were using an unsupported setup -- them using multiple files as a mirror in a VM to test their setup before upgrading the production didn't work...what did work was using the actual disks in the production environment and they were able to narrow that down to an untested use case.

Other issues in that thread are due to how the system initializes and, well, those seem to be systemd bugs and not ZFS bugs because a lot of them are related to services being started too soon/not waiting for dependencies to finish before starting. Read it all from here down.

[skeevy420]

ZFS fanboys are such bizarre creatures. They always, without fail, speak of their filesystem as if it has other-worldly properties that no worldly filesystem can ever hope to replicate.

In reality, ZFS is decidedly mediocre filesystem with some very obvious shortcomings and design flaws. In many ways, btrfs is an improvement over it.

But please, do some more handwaving about "quality" and "features". Don't forget to thank Mr Ellison

Datasets w/o LVM. Functional raid at any level w/o needing to know where there be dragons. Encryption w/o LUKS. ZFS's LZ4 compresses data better than BTRFS using Zstd. Snapshots and more w/o external tools. Case insensitivity as an option. Online deduplication. SELinux support. NFS and SAMBA support. And a shit load more.

There is no other file system available on Linux that offers all of that. Not one.

[lu_tze]

Don't add "not even BTRFS" to that. Don't update your kernel because a new BTRFS bug will destroy your data has happened before and will likely happen again.

Based on reading that page again, it is still mostly user error, distribution startup differences, lack of user knowledge, some outdated documentation, and, in the case of the OG poster, was because they were using an unsupported setup -- them using multiple files as a mirror in a VM to test their setup before upgrading the production didn't work...what did work was using the actual disks in the production environment and they were able to narrow that down to an untested use case.

Other issues in that thread are due to how the system initializes and, well, those seem to be systemd bugs and not ZFS bugs because a lot of them are related to services being started too soon/not waiting for dependencies to finish before starting. Read it all from here down.

The difference is, that btrfs didn't eat my data yet; but zfs did cause several issues already.

This one is just a last in the line; no, its not unsupported configuration with files, like the original poster had (I have a ThinkServe machine with LSI SAS1064ET controller, that exhibits the same); you might note that several different distributions had the same problem, and it is not user configuration error because every unit that is supposed to be enabled is enabled - and yet I always end up in a dracut emergency shell after booting with 0.8.2, because it cannot load its kernel modules from initramfs (yes, they are there).

Since my patience is over, I'm in the process of migrating all the services off this machine and it will be eventually decommissioned. Also, the take-home knowledge is to never use zfs for root file system. Ever.

[skeevy420]

The difference is, that btrfs didn't eat my data yet; but zfs did cause several issues already.

My experiences are the exact opposite of that. I wouldn't trust BTRFS to hold easily re-obtainable porn.

Anyways, it seems like everyone and their Mom has a horror story about *insert Advanced Disk Format here*. Like how ours are opposite of one another with BTRFS and ZFS. I'm sure there is, or will be, someone who reads our discussion and goes "Y'all are some dumb fucks. They both suck. LVM and Ext4 is where it's at".

Yeah, ZFS as root isn't the safest thing to do right now. There's possibly an exception to that for Ubuntu users, but only because that's a supported Ubuntu configuration.

Well, maybe Manjaro users since there's a ZFS on Root option hidden away in the Manjaro-Architect installer...but Manjaro doesn't advertise or promote that and it's just kind of there for people who know what they're doing and want to give it a try...but, IMHO, their default datasets suck ass so it's better to do it manually...

[andrewjoy]

ZFS is a mature file system that has been developed for years and has millions of man hours of work behind it. The feature set is powerful and flexible.

Like any file system tho , RAID is not a backup and you should have robust backups in place. That being said , your data is safer on a properly configured ZFS file system than any other.

Just because it started at sun/oracle and Torvalds does not like it does not make the project bad.

As a root file system , on linux.... here be dragons.