Announcement

Collapse
No announcement yet.

Btrfs To Go Production-Ready In Oracle Linux

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

  • Btrfs To Go Production-Ready In Oracle Linux

    Phoronix: Btrfs To Go Production-Ready In Oracle Linux

    Btrfs, the quite promising next-generation Linux file-system that's been in-development for years by Chris Mason and others, is about to take on a big role within Oracle's Enterprise Linux distribution...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    raid-5

    I'll switch from zfs-fuse when btrfs gets raid-5 support. They said they would do that once fsck was in place. So please implement it!

    Comment


    • #3
      Originally posted by supert0nes View Post
      I'll switch from zfs-fuse when btrfs gets raid-5 support. They said they would do that once fsck was in place. So please implement it!
      If at all possible, you should go with raid 10. That would give you speed advantages as well as increased robustness (as well as enabling data scrub recovery). Raid 10 has also been available for awhile now.
      Raid 5 can be a death trap...

      Comment


      • #4
        Originally posted by supert0nes View Post
        I'll switch from zfs-fuse when btrfs gets raid-5 support. They said they would do that once fsck was in place. So please implement it!
        Right. I played around with zfs-fuse for a little bit. What a horrid mess. Unreliable, SLOOOOOOOOOW. Completely worthless.
        BTW: MDRAID supports raid 5/6.... and is actually reliable. Nothing stopping you from running a btrfs over mdraid.

        Comment


        • #5
          Originally posted by droidhacker View Post
          BTW: MDRAID supports raid 5/6.... and is actually reliable. Nothing stopping you from running a btrfs over mdraid.
          But can it recover from silent corruption errors as btrfs's RAID1(0) implementation could? I imagine, that if RAID5/6 is implemented within btrfs it can recompute the wrong block even if the hardware sees no fault. But with MDRAID the redundancy is transparent and cannot be used to recover, mesa thinks.

          Comment


          • #6
            Michael, you've taken a quote out of context! When Avi said "BtrFS is that fast ALL the time," he was talking about mkfs.btrfs. No matter what size the filesystem is, mkfs will always be that quick. That has no connection to whether an XFS developer was shooting or not.

            Comment


            • #7
              This file system has made great progress over the last couple years, im actually pretty excited about finally seeing it reach a stable enough status to go out with a distro and out to a wider audience. good stuff.

              Comment


              • #8
                Don't hold your breath. It is not working

                Hi,
                The btrfs is not stable when you perform intensive large files saving in parallel threads!!! We were able to crash it twice in raw. Btrfs trying to cache data and save it later. In the result we got a kernel panic. The first time Kernel panic was not recoverable without reboot. The second time computer didn't freeze.
                After this system was completely corrupted and it is not recoverable.

                Code:
                mount -o recovery /dev/vdb /srv/btrfs
                Looks great on video but doesn't working. See below:
                Code:
                # mount -o recovery /dev/vdb /srv/btrfs
                mount: wrong fs type, bad option, bad superblock on /dev/vdb,
                       missing codepage or helper program, or other error
                       In some cases useful info is found in syslog - try
                       dmesg | tail  or so
                #
                Dmesg give this:
                Code:
                device fsid c5ce4702-2dbf-4b57-8067-bd6129fc124b devid 1 transid 1384 /dev/vdb
                btrfs: enabling auto recoveryparent transid verify failed on 20971520 wanted 1347 found 3121
                parent transid verify failed on 20971520 wanted 1347 found 3121
                parent transid verify failed on 20971520 wanted 1347 found 3121
                parent transid verify failed on 29470720 wanted 1357 found 3231
                parent transid verify failed on 29470720 wanted 1357 found 3231
                parent transid verify failed on 29470720 wanted 1357 found 3231
                parent transid verify failed on 29470720 wanted 1357 found 3231
                parent transid verify failed on 29470720 wanted 1357 found 3231
                Failed to read block groups: -5
                btrfs: open_ctree failed
                I think Oracle need to spend couple more years before it will work.

                Comment


                • #9
                  I'm too scared after reading btrfs mailing list...

                  Comment


                  • #10
                    I don't know about Oracle but we have Centos 7 at work that is on kernel 3.10. I've had major issues with BTRFS around 3.18 where a system freeze borked my root partition beyond what the very-lacking fsck for BTRFS would repair. Ended up reinstalling with XFS which is what I was using previously. Subsequently froze the system at least 5 times that night, none of which borked anything with XFS. I know nothing is perfect and things take time to perfect, but even FAT32 didn't have this problem for me. I find it completely unacceptable that a modern filesystem would have that bug. I've mentioned it in other threads one or two other times and some people did not think my reasoning was solid, but I stand behind it. If you're willing to risk uncorrectable corruption then you go ahead and use BTRFS; I'm good.

                    I know distros backport fixes to their old kernels, but there's absolutely no way I'd trust BTRFS in a kernel older than the one I had an issue with unless I knew for a fact that the bug was fixed in that kernel. Again I know distro kernels get fixes but blindly trusting them to have fixed all the issues is not wise in my opinion.

                    Oh and I do use ZFS. I've never benchmarked it under Linux but a 4-drive RAID-10 equivalent was easily able to saturate 1Gbps LAN which is fast enough for the data I put on it. I would easily recommend ZFSonLinux over BTRFS in literally any situation (except maybe to BTRFS developers, as maybe someday they'll figure it out). I would even feel comfortable using ZFSonLinux in production systems.

                    Maybe if Oracle's version of Linux is using some newer kernel like 4.4, I'd look into what bugs have been fixed with BTRFS. Otherwise, heck no it's not production-ready. If I had to remove the conditional there, I'd go with just "heck no it's not production-ready."
                    Last edited by Holograph; 27 January 2017, 02:57 PM.

                    Comment

                    Working...
                    X