[LU-8756] qmt_dqacq0()) $$$ Release too much! and DQACQ failed with -22

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.5.3
• Severity: 3
• Assignee: WC Triage
• Reporter: Andreas Dilger

Description

Most "Release too much" error messages are caused by out of sync between slave and master (several known defects could cause this have been fixed), and such inconsistency can be fixed by quota reintegration, however, this one is different with others I've seen before:

Oct 13 10:09:26 mds01 kernel: : LustreError: 7534:0:(qmt_handler.c:427:qmt_dqacq0()) $$$ Release too much! uuid:lfs_home-MDT0000-lwp-OST0006_UUID release:4 granted:4, total:0 qmt:lfs_home-QMT0000 pool:0-dt id:901898 enforced:1 hard:346030080 soft:314572800 granted:0 time:0 qunit:16777216 edquot:0 may_rel:0 revoke:0

We can see the slave and master are synced, they both think OST0006 owns 4 granted, however, the granted owned by slave doesn't comply with the total granted (which is 0 in this case). On master, the updates of the 'granted owned by each slave' and the 'total granted' are always in pair and in same transaction, so it's supposed that the add up of all slave's granted always equal to the total granted.
I can't think of how this can happen so far, could you open a ticket for further investigation? Thanks a lot.