OpenZFS 2.3-rc4 Released With Linux 6.12 LTS Support

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • varikonniemi
    replied
    I am not neglecting any flaws, i just objectively can see they are much smaller than any other project of similar scale. Meaning it's the best QA out there. And because of it, even if it causes friction and disagreements, even strongly opinionated projects like bcachefs stay grounded and don't come with silent data corruption like what happened in the early days when it was out of tree.

    Go circlejerk your ZFS outdated design in bsd forums or wherever guys that appreciate being corporate drones gather. Here i would appreciate sticking to the facts, like appreciating a completely new design that provides several guarantees. And only one of them is the elegance which results in lower LOC and thus less bugs.

    Leave a comment:


  • mbod
    replied
    Originally posted by varikonniemi View Post

    Don't you think it's expected to be a to-and-forth battle seeing all the insane comments seen here against a truly next-gen filesystem? Only because some fanboys like to cling onto the monstrous piece of out of kernel code instead of looking at the facts. Most people can understand that having several times the amount of code to achieve similar features means there is several times more bugs in there.

    (this applies to btrfs also, it's so huge one cannot come to other conclusion than the design being wrong, and force is used to extend it)
    The only guy not looking at facts here is you.

    You are totally neglecting the flaws in the kernels QA process. You are saying that this is better than anything else out-of-tree. Which is nonsens and shows that you have no background.

    And comparing bcachefs with ZFS is also pointless. bcachefs does not even support scrub yet. Which means that you can not check your data. Not to mention RAID or encryption. bcachefs is just a basic filesystem. That is it. It is a COW filesystem and it happens to have checksums. That is a good start but still a long way to go.

    bcachefs has a very small footprint yet. Not many people use it. The big bugs will be revealed when bcachefs reaches the mainstream and when it supports more sophisticated features like RAID or encryption.

    Also, you should realize that bcachefs is just a one man show. While ZFS is developed by dozens of developers supported by dozens of companies. Your trust in the bcachefs one man show seems odd to me. I would even call it unprofessional.



    Leave a comment:


  • mbod
    replied
    Originally posted by varikonniemi View Post
    and we have already seen it, zfs had a really bad silent data corruption bug, while an experimental bcachefs has had absolutely no bug that resulted in data corruption.

    This is extremely bad for zfs, since it should have shaken out all bugs decades ago, compared to a newcomer that has spotless track record since being merged. While being experimental. Meaning the author is not still completely satisfied with it, but it still has proven better than zfs
    I mentioned it multiple times now: There have been bugs about data corruption in almost any filesystems. ext4, btrfs, xfs, etc.

    The good news about the openzfs bug you are referring too was that it was very hard to trigger. That is the reason why it was discovered and solved so late. You had to write to a file and copy that file while you write to it. Example: copy A -> B and simultaneously cp B -> C while copy process for A is still running. This could lead to data corruption if both copy processes meet certain timing constraints.

    Anyways, you can pick on this one bug. Good for you. But you are just showing that you have no background. Talking about bcachefs and that it has no bugs and that it is super and this is because the developer is a star and the kernel community has the best quality assurance is just ridiculous.

    Leave a comment:


  • mbod
    replied
    Originally posted by varikonniemi View Post

    In tree you have the kernel community assuring the quality. Out of tree you have you assuring it.
    It is not me assuring it. You are funny. I am not assuring nvidia driver quality, I am not assuring vmware quality, I am not assuring openzfs quality. This is all up to the developers. I am following the openzfs development and I am very pleased with what I see there. How they do quality assurance. This is best in class from my point of view. Havent seen it any better anywhere else. And the kernel is certainly not a good example for quality assurance at all. It is too often being release with serious issue. Like kernel 6.12.2 which was relased although it was not booting. Or kernel 6.1.64 and 6.1.65 which had silent data corruption for the ext4 file system: https://lwn.net/Articles/954285/

    Originally posted by varikonniemi View Post
    You can be as activist and zfsist as you want, you won't convince that it's better to use something that has not been vetted by the kernel community compared to if it has been.
    Why do you believe that "the kernel community" is better than any other developer community? The kernel has no real QA process. They have published broken kernels many times in the past. kernels not booting, kernels core dumping, kernels with silent data corruption in ext4, kernels with serious btrfs bugs incl. data corruption, etc. etc.

    Leave a comment:


  • Developer12
    replied
    Originally posted by varikonniemi View Post

    because it has had many. So you are hard to think at any.
    bring receipts.

    Leave a comment:


  • Developer12
    replied
    Originally posted by Chugworth View Post

    I'm not sure what you mean by BPW. They implemented block cloning in 2.2, so they should have most of the building blocks for the type of deduplication that I was talking about.

    Basically I just want to take a dataset, scan it for duplicate blocks in a one-time process (not constantly), then use block cloning to remove the duplicate blocks.
    block cloning is an ENTIRELY different mechanism to the fabled Block Pointer Rewrite. that "scanning" part is hard and "removing" part is basically impossible. once a block in ZFS has been written, that's it. it's basically immutable forever. the only way to get rid of it is to delete or overwrite it at the file level. no internal edits are allowed. thus, all deduplication *must be performed at the time of initial write.*

    Leave a comment:


  • varikonniemi
    replied
    and we have already seen it, zfs had a really bad silent data corruption bug, while an experimental bcachefs has had absolutely no bug that resulted in data corruption.

    This is extremely bad for zfs, since it should have shaken out all bugs decades ago, compared to a newcomer that has spotless track record since being merged. While being experimental. Meaning the author is not still completely satisfied with it, but it still has proven better than zfs
    Last edited by varikonniemi; 16 December 2024, 06:52 PM.

    Leave a comment:


  • varikonniemi
    replied
    Originally posted by mbod View Post

    What do you mean? In-tree quality assurance is the best? Well, you certainly have no clue.

    Is an "internal" developer any better than an "external" developer? Are all these internal developers the cream de la cream of software development with the best in class QA processes? No, they are not. They just offered kernel 6.12.2 which was not booting and had to be fixed with kernel 6.12.3 just one day later. Excellent QA at work. How come that 6.12.2 was even released? It was not tested at all.

    And what about all the regressions related to btrfs and even ext4 in the past? How did they test this stuff? Do you know? No you dont. If you are really interested to learn how state of the art software development works incl. a full featured test suite with test bots testing new features and dozens of known pit falls in ZFS for various operating systems and full transparency about everything you need to check out openzfs on github.
    In tree you have the kernel community assuring the quality. Out of tree you have you assuring it.

    You can be as activist and zfsist as you want, you won't convince that it's better to use something that has not been vetted by the kernel community compared to if it has been.

    Leave a comment:


  • mbod
    replied
    Originally posted by varikonniemi View Post

    It's THE quality assurance. Anything else is quality assurance on behalf of some external developers.
    What do you mean? In-tree quality assurance is the best? Well, you certainly have no clue.

    Is an "internal" developer any better than an "external" developer? Are all these internal developers the cream de la cream of software development with the best in class QA processes? No, they are not. They just offered kernel 6.12.2 which was not booting and had to be fixed with kernel 6.12.3 just one day later. Excellent QA at work. How come that 6.12.2 was even released? It was not tested at all.

    And what about all the regressions related to btrfs and even ext4 in the past? How did they test this stuff? Do you know? No you dont. If you are really interested to learn how state of the art software development works incl. a full featured test suite with test bots testing new features and dozens of known pit falls in ZFS for various operating systems and full transparency about everything you need to check out openzfs on github.

    Leave a comment:


  • varikonniemi
    replied
    Originally posted by NotMine999 View Post
    Oh wunderful. Yet another forum thread living up to the reputation of Moronix with all of the whiny (snarky?) comments about ZFS.
    Don't you think it's expected to be a to-and-forth battle seeing all the insane comments seen here against a truly next-gen filesystem? Only because some fanboys like to cling onto the monstrous piece of out of kernel code instead of looking at the facts. Most people can understand that having several times the amount of code to achieve similar features means there is several times more bugs in there.

    (this applies to btrfs also, it's so huge one cannot come to other conclusion than the design being wrong, and force is used to extend it)

    Leave a comment:

Working...
X