-
Bug
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
3
-
9223372036854775807
conf-sanity test_153c crashed the MDS node (trevis-154vm347, el10.2 ldiskfs, review-dne-part-3, kernel 6.12.0-211.47.1.el10_2_lustre) during cleanup, at the umount of lustre-MDT0001:
LustreError: (obd_config.c:1092:class_decref_free()) ASSERTION( !test_bit(OBDF_ATTACHED, obd->obd_flags) ) failed
LustreError: (obd_config.c:1092:class_decref_free()) LBUG
Workqueue: obd_zombid obd_zombie_imp_cull [obdclass]
class_decref_free+0x98/0xa0 [obdclass]
obd_zombie_import_free+0x249/0x5b0 [obdclass]
process_one_work, worker_thread, kthread
Timeline from the console log: test 153c starts at 40992 s; at 41009 s "lustre-MDT0000-osp-MDT0001: Connection to lustre-MDT0000 was lost"; the test calls cleanup, "umount -d -f /mnt/lustre-mds2" at 41014 s; the zombie import worker hits the assertion at 41020 s. An obd's refcount reached zero while the device was still attached, when the zombie worker freed an import: a refcount imbalance or an obd detached too late on the server umount path.
Seen once in review-dne-part-3: 0 failures of this group in 1362 review runs over the previous 30 days, and 6 of 6 other runs of this change passed. The change under test (LU-20690 EC, lod, lfs and tests) does not touch obdclass, ptlrpc or osp. The only other recent test_153c crash (change 70378, 2026-11-05) is an LNet lnet_assert_handler_unused assertion, a different failure, already linked to LU-20842.
Maloo: https://testing.whamcloud.com/test_sets/bf5abacc-44ae-4650-8c24-3d6361a8fdde
Test log for the subtest: https://testing.whamcloud.com/sub_tests/568150c9-c6a8-4d6d-9885-805e0afa10bb
Tested change: https://review.whamcloud.com/69703 patchset 7
Raised by Patch Watcher run pw-engineer-69703-ps7-72e2573f2e89 on https://review.whamcloud.com/69703
VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
conf-sanity test_153c - trevis-154vm347 crashed during conf-sanity test_153c