-
Bug
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
3
-
9223372036854775807
sanity-quota test_73b ("default OST Pool Quotas limit for new user") hangs and the sanity-quota test set ends with "Timeout occurred after N minutes, last suite running was sanity-quota". It is a deadlock in the quota pool code on the MDS, not a data path failure.
The suite_log stops right after "set default quota for qpool1", i.e. at the runas dd in test_73b. The MDS stack_trace log from the timed out subtest shows about 55 mdt_rdpg threads blocked in rwsem_down_read_slowpath from qti_pools_add, one mdt thread blocked in rwsem_down_read_slowpath (cfs_hash_lookup / cfs_hash_multi_bd_lock), and qsd_reint_qpool blocked in rwsem_down_write_slowpath (cfs_hash_find_or_add / cfs_hash_multi_bd_lock). Clients are evicted/lose connection to MDT0000 while the hang lasts.
Frequency: rainshadow test sanity-quota test_73b shows 121 branch failures in 30 days (lustre-master-next 65, lustre-master 56), all TIMEOUT, first seen 2026-11-02, in full-part-2 (about 13-15% of runs) and full-dne-part-2 (17-25%). Reviews ran 6604 suite runs with a single TIMEOUT, on change 64875 (LU-19783 lfsck, unrelated: the patch does not touch quota or the default pool quota path).
Example failures:
master full-part-2 (same stack): https://testing.whamcloud.com/sub_tests/a3490105-320a-441d-b648-88701ce9807a
review-zfs on change 64875 ps39: https://testing.whamcloud.com/sub_tests/51859b9d-cc9b-42fe-809b-cc42995966c9
No existing ticket matches (LU-16448, LU-17974 are qsd_reint_qpool crashes, resolved). Likely a lock ordering problem between qsd_reint_qpool and qti_pools_add/lqe pool lookups (see the nested lqe lookup fix for test_73b).
Raised by Patch Watcher run pw-engineer-64875-ps39-517a61d8daf3 on https://review.whamcloud.com/64875
VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
sanity-quota test_73b - Timeout occurred after 939 minutes, last suite running was sanity-quota