Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-20526

lod_ost_alloc_qos does not handle offline OSTs well

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Medium Medium
    • None
    • None
    • None
    • 3
    • 9223372036854775807

      The tracking of the total_weight seems out of sync between the initial loop where it is computed and the stripe allocator loop, often causing it to fail to find an OST due to the weight miscompute.

      This should be fixed.

      /* TODO: this check is wrong and can cause the qos

      • allocator to fail.the problem is that we do not
      • compute total_weight and cur_weight the same way
      • when OSTs are offline.
      • Very annoying since if OSTs are offline this will
      • fail, and fallback to rr quite often, making
      • testing painful.
        */

            adilger Andreas Dilger
            rsahlberg@whamcloud.com Ronnie Sahlberg
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated: