Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-21233

lfs: report the volatile FID in mirror extend/migrate --stats output

XMLWordPrintable

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

      lfs mirror extend, lfs migrate and lfs migrate -m-to-DoM copy the data through an open-unlinked volatile file. The new layout is then merged or swapped into the original file. This volatile file has its own FID. It has no name, so lfs fid2path can't resolve it, and it is gone once the command finishes.

      As a result, the OSTs record every write the copy makes under a FID that nothing else ever reports. Tools that watch OST access (for example readers of the OFD access log, or tiering and heat tracking) see a large burst of writes to an unknown FID. They can't tell this I/O apart from application I/O.

      --stats already prints a YAML progress stream for whoever runs the copy.
      Proposal: print the volatile FID once, right after the file-name header and before any data is copied:{}

      /mnt/lustre/dir/file:
      - { volatile_fid: "[0x200000401:0x2:0x0]" }
      - { seconds: 1, rmbps: 16, wmbps: 16, copied: 16, total: 32, pct: 50% }

       

            npapakonstantinou Nikos Papakonstantinou
            npapakonstantinou Nikos Papakonstantinou
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

              Created:
              Updated: