Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-21203

osd-zfs: detect non-rotational VDEVs at mount

XMLWordPrintable

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

      osd-ldiskfs sets the target's nonrotational flag (OS_STATFS_NONROT, added in LU-11963) from its block device at mount, but osd-zfs never sets it unless the administrator sets osd-zfs.*.nonrotational=1 by hand. A ZFS OST on SSD or NVMe devices is therefore treated like one on disks, e.g. by FLR mirror selection on the client and MDS.

      ZFS already tracks this per vdev in vdev_nonrot: a disk vdev takes it from blk_queue_nonrot() of its block device, a file vdev is always non-rotational, and a mirror, RAID-Z or dRAID vdev is non-rotational when all of its children are.

      osd-zfs can detect it at mount by walking the top-level vdevs of the normal allocation class under spa_config_enter(SCL_VDEV), skipping log, special, dedup, hole and removed (indirect) vdevs, and setting od_nonrotational when all of them are non-rotational. The nonrotational parameter stays as an override.

            wc-triage WC Triage
            adilger Andreas Dilger
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

              Created:
              Updated: