Details

    • Bug
    • Resolution: Fixed
    • Minor
    • Lustre 2.8.0
    • None
    • None
    • 3
    • 15223

    Description

      This ticket is used to track enhancements that appear to be required during Lustre DKMS RPM build/testing integration in our build/tests infrastructures

      This is a follow-on to LU-1032 main ticket that introduced DKMS usage/need for Lustre.

      Attachments

        Issue Links

          Activity

            [LU-5465] Enhancements to Lustre DKMS RPM

            The patches needed for this work landed in 2.8

            jgmitter Joseph Gmitter (Inactive) added a comment - The patches needed for this work landed in 2.8

            Oleg Drokin (oleg.drokin@intel.com) merged in patch http://review.whamcloud.com/11776/
            Subject: LU-5465 build: strengthen Lustre DKMS RPM install
            Project: fs/lustre-release
            Branch: master
            Current Patch Set:
            Commit: 7d31575a5e2a488c289ce17a2d7a766d8856adf4

            gerrit Gerrit Updater added a comment - Oleg Drokin (oleg.drokin@intel.com) merged in patch http://review.whamcloud.com/11776/ Subject: LU-5465 build: strengthen Lustre DKMS RPM install Project: fs/lustre-release Branch: master Current Patch Set: Commit: 7d31575a5e2a488c289ce17a2d7a766d8856adf4

            Oops, you are right, I have been to quick on this ... Will re-close LU-5104 and work on this in LU-4606!

            bfaccini Bruno Faccini (Inactive) added a comment - Oops, you are right, I have been to quick on this ... Will re-close LU-5104 and work on this in LU-4606 !

            bfaccini: Are you sure about that? Change 10654 only wraps the:

            %{_libdir}/@PACKAGE@/mount_osd_ldiskfs.so
            

            that's already in the osd-ldiskfs with a:

            %if %{with lustre_utils}
            ...
            %endif
            

            mount_osd_ldiskfs.so seems to have actually been added to the osd-ldiskfs package by LU-4606 in change 10193 from what I can see.

            Maybe what needs to happen here is that lustre-osd-ldiskfs needs to be split into lustre-osd-ldiskfs and kmod-osd-ldiskfs (and of course the same for the ZFS osd module). Then lustre-dkms can provide/conflicts the kmod-osd zfs and ldiskfs packages and the lustre-osd zfs and ldiskfs packages can continue to provide the userspace hooks.

            brian Brian Murrell (Inactive) added a comment - bfaccini : Are you sure about that? Change 10654 only wraps the: %{_libdir}/@PACKAGE@/mount_osd_ldiskfs.so that's already in the osd-ldiskfs with a: % if %{with lustre_utils} ... %endif mount_osd_ldiskfs.so seems to have actually been added to the osd-ldiskfs package by LU-4606 in change 10193 from what I can see. Maybe what needs to happen here is that lustre-osd-ldiskfs needs to be split into lustre-osd-ldiskfs and kmod-osd-ldiskfs (and of course the same for the ZFS osd module). Then lustre-dkms can provide/conflicts the kmod-osd zfs and ldiskfs packages and the lustre-osd zfs and ldiskfs packages can continue to provide the userspace hooks.

            I have re-opened LU-5104 to point to this functionality/compatibility problem between both tickets/changes.

            bfaccini Bruno Faccini (Inactive) added a comment - I have re-opened LU-5104 to point to this functionality/compatibility problem between both tickets/changes.

            Brian: I was not aware of the fact that mount_osd_[zfs,ldiskfs].so libs are now part of lustre-osd-[zfs,ldiskfs] RPMs, but this seems to come from a recently integrated fix (Gerrit Change #10654, with commit adf2e9d9ab987b46fa0269038f348cf2a17fd0e0, for LU-5104). In fact, I don't know presently how we can fix this, because of LU-2391 osd_[zfs,ldiskfs].ko have been packaged in their own RPMs, and at that time they were not expected to contain something else than back-end specific Kernel modules ...

            bfaccini Bruno Faccini (Inactive) added a comment - Brian: I was not aware of the fact that mount_osd_ [zfs,ldiskfs] .so libs are now part of lustre-osd- [zfs,ldiskfs] RPMs, but this seems to come from a recently integrated fix (Gerrit Change #10654, with commit adf2e9d9ab987b46fa0269038f348cf2a17fd0e0, for LU-5104 ). In fact, I don't know presently how we can fix this, because of LU-2391 osd_ [zfs,ldiskfs] .ko have been packaged in their own RPMs, and at that time they were not expected to contain something else than back-end specific Kernel modules ...

            bfaccini: With http://review.whamcloud.com/11776, shouldn't the osd_zfs.ko module also be removed from the lustre-osd-zfs RPM? Is that being tracked somewhere else?

            brian Brian Murrell (Inactive) added a comment - bfaccini : With http://review.whamcloud.com/11776 , shouldn't the osd_zfs.ko module also be removed from the lustre-osd-zfs RPM? Is that being tracked somewhere else?

            2nd patch is at http://review.whamcloud.com/11776, to strengthen Lustre DKMS RPM install vs legacy lustre-osd and lustre-modules RPMs already installed, with more Provides/Conflicts statements spec file.

            bfaccini Bruno Faccini (Inactive) added a comment - 2nd patch is at http://review.whamcloud.com/11776 , to strengthen Lustre DKMS RPM install vs legacy lustre-osd and lustre-modules RPMs already installed, with more Provides/Conflicts statements spec file.

            1st patch is at http://review.whamcloud.com/11381, to address case where SPL/ZFS have only been "added" (vs "build" and/or "install") in DKMS terminology and be able to dynamically find their respective version anyway in Lustre DKMS RPM dkms.conf script. This should strengthen Lustre DKMS install and automatic build.

            bfaccini Bruno Faccini (Inactive) added a comment - 1st patch is at http://review.whamcloud.com/11381 , to address case where SPL/ZFS have only been "added" (vs "build" and/or "install") in DKMS terminology and be able to dynamically find their respective version anyway in Lustre DKMS RPM dkms.conf script. This should strengthen Lustre DKMS install and automatic build.

            People

              bfaccini Bruno Faccini (Inactive)
              bfaccini Bruno Faccini (Inactive)
              Votes:
              0 Vote for this issue
              Watchers:
              11 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: