[LU-2153] Quota reintegration holes

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.4.0
• Affects Version/s: Lustre 2.4.0
• Severity: 3
• Assignee: Niu Yawei (Inactive)
• Reporter: Johann Lombardi (Inactive)

Description

While testing quota with ZFS, i found a couple of issues in the reintegration procedure that would be great to fix:

• we remove the entry from global index once all limits are set to 0 and granted drops to 0. However, we might still have some disconnected OSTs with enforced bit set and usage = 0 for this ID which won't be notified during reintegration that quota isn't enforced any more for this ID. We should probably modify test 7a to test this case.
• the reintegration thread can run at the same time as the writeback thread (which might process some old record updates which have been delayed for whatever reason). We need to review the locking to make sure that the writeback thread cannot corrupt the slave index behind our back.
• reintegration can in theory happen at any time, so we need to make sure (via code review first) that we can't race with any other actions in progress (e.g. dqacq request in flight ...). In review 4253, i added a proc file to trigger reintegration on demand, we can use this parameter to add more tests if needed.

Thanks.