Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-21079

sanity test_160h: 'mds1: first index should be N + 1, but is N' orphan changelog record not deleted at mount

XMLWordPrintable

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

      sanity test_160h (changelog gc thread stop upon umount, orphan records delete during mount) failed in review-dne-zfs-part-1 (4 MDTs on 2 MDS, ZFS, el9.8) with:

      mds1: first index should be 34 + 1, but is 34

      The check is sanity.sh test_160h after the MDTs are restarted: cl_user1 (cl10) was cleared to record #34 while the GC thread was held by OBD_FAIL_FORCE_GC_THREAD, and the first record left on MDT0000 is still #34. The orphan record that should have been deleted at mount, past cl_user2's (cl11) removal, was not.

      This is not the 'User clN not registered' (LU-15566), 'expected N GC-thread' (LU-20884) or 'User clN still registered' (LU-19933) failure of the same test: all users were in the expected state, only the first record index is off by one. Rainshadow shows 0 occurrences of this error in 730 days, and the failure is a one-off in 30 days for this error.

      Failing run: https://testing.whamcloud.com/sub_tests/efe68910-5da4-4bc4-bb9e-dbde1ed99c82
      Seen on https://review.whamcloud.com/69392 (an EC-only patch that does not touch changelog code).

      Raised by Patch Watcher run pw-engineer-69392-ps3-7c1bae325c3f on https://review.whamcloud.com/69392

      VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
      sanity test_160h - mds1: first index should be 34 + 1, but is 34

            paf0186 Patrick Farrell
            maloo Maloo
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated:
              Resolved: