Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-5923

Master testing: MDT0 not mounting after master downgrade to 2.5

Details

    • Bug
    • Resolution: Duplicate
    • Critical
    • None
    • None
    • None
    • File system running 2.5.2 (CentOS 6.5 servers), started with master (most recent commit Ie7a2a98be8cc97db9af7a64476c06fc7321544eb, http://review.whamcloud.com/12142), went back to 2.5.2.
    • 3
    • 16532

    Description

      After moving a file system running 2.5.2 to master and then moving back, we are unable to mount MDT0:

      LustreError: 3262:0:(lfsck_namespace.c:147:lfsck_namespace_load()) perses1-MDT0000-: fail to load lfsck_namespace, expected = 256, rc = -34
      LustreError: 3262:0:(mdd_device.c:970:mdd_prepare()) perses1-MDD0000: failed to initialize lfsck: rc = -34
      LustreError: 3262:0:(mdt_handler.c:6264:mdt_iocontrol()) perses1-MDT0000: Aborting recovery for device
      LustreError: 3262:0:(obd_mount_server.c:1739:server_fill_super()) Unable to start targets: -34

      I've already tried a write_conf, which didn't change anything.

      If anyone has an idea of the root cause or how to resolve this for the active file system, I'd appreciate the input, as it's down at the moment.

      Attachments

        Issue Links

          Activity

            [LU-5923] Master testing: MDT0 not mounting after master downgrade to 2.5

            Thanks, Andreas - As you saw, I figured out the workaround right before you posted. I appreciate the quick response. I'll also review that patch...

            paf Patrick Farrell (Inactive) added a comment - Thanks, Andreas - As you saw, I figured out the workaround right before you posted. I appreciate the quick response. I'll also review that patch...

            The patch http://review.whamcloud.com/12321 from LU-5707 should fix this bug. A separate patch to b2_5 will allow it to mount the filesystem even if there is a problem loading this file.

            adilger Andreas Dilger added a comment - The patch http://review.whamcloud.com/12321 from LU-5707 should fix this bug. A separate patch to b2_5 will allow it to mount the filesystem even if there is a problem loading this file.

            Mount the MDT as type ldiskfs and delete the lfsck_namespace file. This is a bug in the current master code that is being fixed before release.

            adilger Andreas Dilger added a comment - Mount the MDT as type ldiskfs and delete the lfsck_namespace file. This is a bug in the current master code that is being fixed before release.

            lfsck_namespace file from the affected system. Removing this file fixed the -ERANGE error.

            paf Patrick Farrell (Inactive) added a comment - lfsck_namespace file from the affected system. Removing this file fixed the -ERANGE error.

            It looks like the ERANGE is bubbling up out of xattr.c in ldiskfs, which I find worrying.

            I was able to get the file system to start again by mounting the MDT as ldiskfs and removing the 'lfsck_namespace' file. I'll attach that file here.

            paf Patrick Farrell (Inactive) added a comment - It looks like the ERANGE is bubbling up out of xattr.c in ldiskfs, which I find worrying. I was able to get the file system to start again by mounting the MDT as ldiskfs and removing the 'lfsck_namespace' file. I'll attach that file here.

            People

              wc-triage WC Triage
              paf Patrick Farrell (Inactive)
              Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: