5 ms·

I would not trust in the quality of any ZFS features in any implementation that were not developed by Sun or at least thought out by Sun before it was acquired.

BoardsOfCanada 4y ago
Care to elaborate?
- Avamander 4y ago
  ZFS on Linux is very buggy. And when something breaks (maybe because one of those numerous bugs) there are no proper tools for repair. Your ZRAID and ZLOG probably won't matter. The to-go tool for removing corrupt superblocks is a 10 year old hacky python script, it's ridiculous.
  - mustache_kimono 4y ago
    Not been my experience at all. I have seen folks get in over their heads though. Like -- "OMG I've somehow broken boot for ZFS on root and I don't know how to fix!!" or "I'm using a custom kernel and OMG things don't work!!" Canonical has even been complicit. When you download an HWE kernel, you don't also download a newer/matching versions of the tools. This should pretty clearly be considered a bug[0], and fixed but Canonical perhaps doesn't have the bandwidth? I just can't lay any of these problems at the feet of ZFS though. Right now I trust ZFS a hell of a lot more than I trust Linux to Do The Right Thing. When ZFS seems broken (though I have little experience with native encryption so this may not apply to it), I am far more likely to think the issue is Linux, my distribution, or me, than any problem with ZFS. I'm a pretty serious ZFS lover though.[1] [0]: https://bugs.launchpad.net/ubuntu/+source/zfs-linux/+bug/1939210 https://bugs.launchpad.net/ubuntu/+source/zfs-linux/+bug/193... [1]: https://github.com/kimono-koans/httm https://github.com/kimono-koans/httm
  - SkyMarshal 4y ago
    Not my experience. I've been running natively-encrypted ZFS on Linux for over a year now with zero bugs or errors. It may help that my distro NixOS has excellent support for it.
    - Avamander 4y ago
      Only a year? I've used ext4 for more than seven years and ZFS for a few months. ZFS shat itself irrecoverably - ZRAID, ZLOG (PLP), ECC RAM and UPS be damned. People's issues sit open on GitHub for years. ZFS on Linux is empirically and anecdotally not trustworthy.
  - mgerdts 4y ago
    The last time I needed to use fsck to repair a filesystem was less than six months ago. After running it I found that my entire directory hierarchy was garbage and lost+found had at least hundreds of files named by their inode numbers. It would have taken me forever to piece this back together if it was a filesystem that I cared about. This is consistent with my experience with fsck in decades past. In 15 years of using zfs, I’ve never lost a pool when it was in a mirror or raidz layout.
- freemint 4y ago
  Sun had certain values as a company it enforced through its corporate culture which made ZFS excellent. My perception is that the ZoL based fork had a lot more instability and bugs compared to the parallel development of ZFS by in the Open Source Solaris world. That OpenZFS is based on the more unreliable code base should be evidence that the values which made ZFS excellent under Suns steermanship are insufficiently maintained and that features integrated into ZFS after the fork who did come from the ZoL side should not be used if one wants the reliability of ZFS. The fact that not all metadata is encrypted shows they are doing things wrong from a principled standpoint. Sun didn't halfbake important stuff like that. The ZoL community will in the long run make the reliability of OpenZFS converge against the reliability of BTRFS because that is the level of reliability open source allows for. Initially higher stability be dammed.
  - mgerdts 4y ago
    Many of the open source zfs developers are the same ones that developed zfs at Sun and then Oracle. I worked on the Solaris team from 2010 until 2017 and have contributed to open source zfs implantations in illumos and openzfs. Your criticisms seem ill-informed based on my actual experience in my contributions to these multiple zfs code bases. I actually found a fair number of improvements in the testing in ZoL. With Solaris zfs, you saw features after they were exercised by up to a few hundred engineers for months to years. Now you are getting bits fresh from the factory. Like when software is developed behind closed doors, bugs happen and they are progressively squashed leading to a more stable product.
  - mgerdts 4y ago
    > The fact that not all metadata is encrypted This was one of my concerns when I first encountered this trade off in Solaris zfs. That is, you claim that the Solaris zfs encryption is better in this regard but it’s not.
    - mustache_kimono 4y ago
      What metadata is not encrypted? I thought it was only dataset names and their properties, which kind of makes sense?
panick21_ 4y ago
Not sure where you are getting that. Lots of former Sun people now work on the OpenSource Solaris. Matt Ahrens was one was the Nr.2 guy who did most of the implementation and he is still heavily involved in OpenZFS. OpenZFS has a very healthy and active community working on it. And the code has been much improved compared to originally open sourced version by Sun. Technology has moved on since then and so has ZFS.

BoardsOfCanada 4y ago

Care to elaborate?

Avamander 4y ago
ZFS on Linux is very buggy. And when something breaks (maybe because one of those numerous bugs) there are no proper tools for repair. Your ZRAID and ZLOG probably won't matter. The to-go tool for removing corrupt superblocks is a 10 year old hacky python script, it's ridiculous.
- mustache_kimono 4y ago
  Not been my experience at all. I have seen folks get in over their heads though. Like -- "OMG I've somehow broken boot for ZFS on root and I don't know how to fix!!" or "I'm using a custom kernel and OMG things don't work!!" Canonical has even been complicit. When you download an HWE kernel, you don't also download a newer/matching versions of the tools. This should pretty clearly be considered a bug[0], and fixed but Canonical perhaps doesn't have the bandwidth? I just can't lay any of these problems at the feet of ZFS though. Right now I trust ZFS a hell of a lot more than I trust Linux to Do The Right Thing. When ZFS seems broken (though I have little experience with native encryption so this may not apply to it), I am far more likely to think the issue is Linux, my distribution, or me, than any problem with ZFS. I'm a pretty serious ZFS lover though.[1] [0]: https://bugs.launchpad.net/ubuntu/+source/zfs-linux/+bug/1939210 https://bugs.launchpad.net/ubuntu/+source/zfs-linux/+bug/193... [1]: https://github.com/kimono-koans/httm https://github.com/kimono-koans/httm
- SkyMarshal 4y ago
  Not my experience. I've been running natively-encrypted ZFS on Linux for over a year now with zero bugs or errors. It may help that my distro NixOS has excellent support for it.
  - Avamander 4y ago
    Only a year? I've used ext4 for more than seven years and ZFS for a few months. ZFS shat itself irrecoverably - ZRAID, ZLOG (PLP), ECC RAM and UPS be damned. People's issues sit open on GitHub for years. ZFS on Linux is empirically and anecdotally not trustworthy.
- mgerdts 4y ago
  The last time I needed to use fsck to repair a filesystem was less than six months ago. After running it I found that my entire directory hierarchy was garbage and lost+found had at least hundreds of files named by their inode numbers. It would have taken me forever to piece this back together if it was a filesystem that I cared about. This is consistent with my experience with fsck in decades past. In 15 years of using zfs, I’ve never lost a pool when it was in a mirror or raidz layout.
freemint 4y ago
Sun had certain values as a company it enforced through its corporate culture which made ZFS excellent. My perception is that the ZoL based fork had a lot more instability and bugs compared to the parallel development of ZFS by in the Open Source Solaris world. That OpenZFS is based on the more unreliable code base should be evidence that the values which made ZFS excellent under Suns steermanship are insufficiently maintained and that features integrated into ZFS after the fork who did come from the ZoL side should not be used if one wants the reliability of ZFS. The fact that not all metadata is encrypted shows they are doing things wrong from a principled standpoint. Sun didn't halfbake important stuff like that. The ZoL community will in the long run make the reliability of OpenZFS converge against the reliability of BTRFS because that is the level of reliability open source allows for. Initially higher stability be dammed.
- mgerdts 4y ago
  Many of the open source zfs developers are the same ones that developed zfs at Sun and then Oracle. I worked on the Solaris team from 2010 until 2017 and have contributed to open source zfs implantations in illumos and openzfs. Your criticisms seem ill-informed based on my actual experience in my contributions to these multiple zfs code bases. I actually found a fair number of improvements in the testing in ZoL. With Solaris zfs, you saw features after they were exercised by up to a few hundred engineers for months to years. Now you are getting bits fresh from the factory. Like when software is developed behind closed doors, bugs happen and they are progressively squashed leading to a more stable product.
- mgerdts 4y ago
  > The fact that not all metadata is encrypted This was one of my concerns when I first encountered this trade off in Solaris zfs. That is, you claim that the Solaris zfs encryption is better in this regard but it’s not.
  - mustache_kimono 4y ago
    What metadata is not encrypted? I thought it was only dataset names and their properties, which kind of makes sense?

panick21_ 4y ago

Not sure where you are getting that. Lots of former Sun people now work on the OpenSource Solaris. Matt Ahrens was one was the Nr.2 guy who did most of the implementation and he is still heavily involved in OpenZFS. OpenZFS has a very healthy and active community working on it. And the code has been much improved compared to originally open sourced version by Sun. Technology has moved on since then and so has ZFS.