Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-20901

sanity test_0e test_3: EIO after spurious MDT0000-osp-MDT0002 eviction at startup

XMLWordPrintable

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

      sanity test_0e and test_3 both FAIL with "Input/output error" moments
      after the review-dne-selinux-ssk-part-1 cluster finishes setup.

      trevis-156vm29 (hosting both lustre-MDT0000 and lustre-MDT0002) logs,
      at the exact second test_0e starts:

      Lustre: lustre-MDT0000: Received MDS connection from 10.240.46.120@tcp,
      removing former export from 0@lo
      Lustre: lustre-MDT0000-lwp-MDT0002: Connection to lustre-MDT0000
      (at 0@lo) was lost; in progress operations using this service will
      wait for recovery to complete
      LustreError: lustre-MDT0000-osp-MDT0002: This client was evicted by
      lustre-MDT0000; in progress operations using this service will fail.

      followed moments later by:

      Lustre: lustre-MDT0000: already connected client
      lustre-MDT0000-lwp-OST0000_UUID (at 10.240.46.130@tcp) with handle
      ... Rejecting client with the same UUID trying to reconnect with
      handle ...

      test_0e's rmdir and test_3's "lfs mkdir -i3" both need the co-located
      MDT0000/MDT0002 OSP, which this eviction just dropped, and both return
      EIO in that window.

      The eviction path is ldlm_lib.c's generic "removing former export"
      connection-replace logic when a client reconnects from a different NID
      than its existing export used; it is not part of this patch's diff
      (lustre/mgs, lustre/target/tgt_mount.c and lustre/mgc changes on this
      change are about MGS-bit reporting over nidtbl, not export or LWP/OSP
      connection handling). test_0e and test_3 both have sporadic failures
      on unrelated changes in the last 30 days with no shared feature, which
      points to a pre-existing startup race when several targets mount and
      reconnect in quick succession on a shared node, not something this
      patch introduces.

      Maloo session: https://testing.whamcloud.com/test_sessions/1833010c-bd1f-4fed-a786-e11c81cb855f
      Test set: https://testing.whamcloud.com/test_sets/2c251607-6aa8-4c8f-9b1f-debff4d06b7e

      Raised by Patch Watcher run pw-engineer-68474-ps10-dfc3aca5a570 on https://review.whamcloud.com/68474

            wc-triage WC Triage
            maloo Maloo
            Votes:
            0 Vote for this issue
            Watchers:
            0 Start watching this issue

              Created:
              Updated: