Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-20183

KUC (and HSM) do not handle short-read

    XMLWordPrintable

Details

    • Bug
    • Resolution: Fixed
    • Medium
    • Lustre 2.18.0
    • None
    • None
    • 3
    • 9223372036854775807

    Description

      When using HSM, it is possible for some coordinator workload to write
      KUC messages larger than a page, by example when sending an HSM request with multiple entries with large 'data' field in it. The total request size could be larger than a 4k-page.

      libcfs_ukuc_msg_get() is calling read() and is expecting this read to get a full kuc message in 1 call. However, when the message is larger than a page, and the system is under load, it is possible that the read() will return the 1st page first, and a second call is needed to get the full page. This is causing short read, and return -EPROTO, and message drop.

      Attachments

        Activity

          People

            adegremont_nvda Aurelien Degremont
            adegremont_nvda Aurelien Degremont
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: