Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-1382

Quota are broken by filesystem extension

Details

    • Bug
    • Resolution: Fixed
    • Minor
    • None
    • Lustre 2.1.0, Lustre 2.1.1
    • None
    • 3
    • 6405

    Description

      After a filesystem extension (ie add of new OSTs), quota are not propagated on newer OSTs.
      Running quotacheck, quotaoff or quotaon doesn't change anything to the picture.

      Here is the output of lfs quota on a fs where OST 0x64..0x80 have been freshly added :

      # lfs quota -v -u foo /scratch/
      Disk quotas for user foo (uid 1000):
          Filesystem kbytes quota limit grace files quota limit grace
       /scratch/ 6985760 21474836480 22548578304 - 8 2000000 2100000 -
      scratch-MDT0000_UUID
                          12 - 131072 - 8 - 5120 -
      scratch-OST0000_UUID
                           0 - 131072 - - - - -
      
      [...]
      
      scratch-OST0062_UUID
                           0 - 131072 - - - - -
      scratch-OST0063_UUID
                     1425412 - 1572864 - - - - -
      scratch-OST0064_UUID
                           0 - 131072 - - - - -
      scratch-OST0065_UUID
                           0 - 131072 - - - - -
      
      [...]
      
      scratch-OST0076_UUID
                           0 - 131072 - - - - -
      scratch-OST0077_UUID
                           0 - 131072 - - - - -
      scratch-OST0078_UUID
                           0 - 0 - - - - -
      scratch-OST0079_UUID
                           0 - 0 - - - - -
      scratch-OST007d_UUID
                           0 - 0 - - - - -
      scratch-OST007e_UUID
                           0 - 0 - - - - -
      scratch-OST007f_UUID
                      677892 - 0 - - - - -
      scratch-OST0080_UUID
                           0 - 0 - - - - -
      

      The side effect is that (most of the) newly created files are not enforced as fas as they are (mostly) created on the new OSTs (the accounting looks good so far).
      The workaround we have consist of removing quota (-b 0 -B 0 -i 0 -I 0) then re-applying them, user by user, but it would be better if quota were automatically propagated during add of new OSTs or, at least, handled by a quotacheck.

      Thanks,

      Attachments

        Activity

          [LU-1382] Quota are broken by filesystem extension

          Have opened LUDOC-65 to track the manual updates. Closing this issue.

          cliffw Cliff White (Inactive) added a comment - Have opened LUDOC-65 to track the manual updates. Closing this issue.
          pjones Peter Jones added a comment -

          Cliff

          Could you please look into this manual update?

          Thanks

          Peter

          pjones Peter Jones added a comment - Cliff Could you please look into this manual update? Thanks Peter

          Ok. In this case it would be a good idea to add a paragraph into the documentation about the workaround before the real fix comes.

          louveta Alexandre Louvet (Inactive) added a comment - Ok. In this case it would be a good idea to add a paragraph into the documentation about the workaround before the real fix comes.

          yes, the only way for now is to re-set quota for each user. The new quota we are working on can address such problem.

          niu Niu Yawei (Inactive) added a comment - yes, the only way for now is to re-set quota for each user. The new quota we are working on can address such problem.
          pjones Peter Jones added a comment -

          Niu

          Could you please look into this one?

          Thanks

          Peter

          pjones Peter Jones added a comment - Niu Could you please look into this one? Thanks Peter

          > The workaround we have consist of removing quota (-b 0 -B 0 -i 0 -I 0) then re-applying them,
          > user by user, but it would be better if quota were automatically propagated during add of
          > new OSTs or, at least, handled by a quotacheck.

          I'm afaid that this is the only solution for now. The new quota architecture will address this problem.

          johann Johann Lombardi (Inactive) added a comment - > The workaround we have consist of removing quota (-b 0 -B 0 -i 0 -I 0) then re-applying them, > user by user, but it would be better if quota were automatically propagated during add of > new OSTs or, at least, handled by a quotacheck. I'm afaid that this is the only solution for now. The new quota architecture will address this problem.

          People

            cliffw Cliff White (Inactive)
            louveta Alexandre Louvet (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: