Details
-
Bug
-
Resolution: Cannot Reproduce
-
Minor
-
None
-
Lustre 2.7.0
-
3
-
15890
Description
There is race between between starting an target (MDT, OST .. both may have it bug) and MGC thread to reenqueue lock.
Target called MGC to connect to MGS to take an config but due to network issue/s config request fails. Because of this target have rollback to the reading config from disk and start re-enqueue thread to take config from network. Target finished read config from a disk and disconnected disk from MGC. But re-euqueue thread still using it. This resulted in a case where thread have an access to the freed structure and panic.
This issue is not present in master code. But test which can created this problem is missing. So adding the test.