Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-19959

quota: code refactoring around qpi_slv_nr

XMLWordPrintable

    • Icon: Task Task
    • Resolution: Unresolved
    • Icon: Minor Minor
    • None
    • None
    • None
    • 3
    • 9223372036854775807

      It seems qpi_slv_nr is not needed anymore as this counter is duplicated by qpi_sarr.osts.op_count(see qmt_sarr_count).

      qpi_slv_nr is incremented when QMT registers new target(OST or MDT). At the same time QMT adds this target at lu_tgt_pool that has it's own counter - osts.op_count(don't care about the name osts - at the QMT level osts is used for MDTs as well). Supporting both counters makes code more complicated adding redundancy.

      The most obvious way is to remove qpi_slv_nr proceeding with osts.op_count(qmt_sarr_count) instead. The drawback of removing qpi_slv_nr is that won't know the exact number of online targets. Finally, qpi_slv_nr is used to calculate qunit. It looks more fair to take into account only online OSTs(and MDTs), not all registered. On the other hand, how often do we have registered but offline targets? Even if some of targets failed, it would appear online soon. At first look, we can live without the accurate number of online targets. At least from the moment when PQs were introduced qpi_slv_nr doesn't show the exact number of online targets. Thoughts?

            scherementsev Sergey Cheremencev
            scherementsev Sergey Cheremencev
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

              Created:
              Updated: