wrong stop action rc when a target is mounted twice
Closed this issue · 3 comments
degremont commented
When a target is also mounted as ldiskfs, a shine stop action to this target leads to incoherent return code (0), the ldiskfs mount point is dismounted but the lustre one is still online (stop output result is correct in that sense).
In case of multi-mounted target, we should detect it, either: dismount all associated mount points (probably the best way?), or at last return an unexpected rc for stop as the lustre target is still mounted.
[root@node1 ~] # shine status -f bwork
= FILESYSTEM STATUS (bwork) =
TYPE # STATUS NODES
---- - ------ -----
MGT 1 online node115
MDT 1 online node115
OST 48 online node[124-127]
CLI 1 mounted node12
[root@node1 ~] # ssh node115
[root@node115 ~] # df
/dev/sdb1 54420 6924 44688 14% /mnt/bwork/mgt
/dev/sdb2 365364336 36043708 304958460 11% /mnt/bwork/mdt/0
[root@node115 ~] # mkdir /mnt/ldiskfs
[root@node115 ~] # mount -t ldiskfs /dev/sdb1 /mnt/ldiskfs
[root@node115 ~] # df
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/sdb1 54420 6924 44688 14% /mnt/bwork/mgt
/dev/sdb2 365364336 36043708 304958460 11% /mnt/bwork/mdt/0
/dev/sdb1 54420 6924 44688 14% /mnt/ldiskfs
[root@node115 ~] # shine status -f bwork -t mgt -L
Status of MGS (/dev/sdb1) ...
Status of MGS (/dev/sdb1) succeeded
= FILESYSTEM STATUS (bwork) =
TYPE # STATUS NODES
---- - ------ -----
MGT 1 online node115
[root@node115 ~] # shine stop -f bwork -t mgt -L
Stop of MGS (/dev/sdb1) ...
Stop of MGS (/dev/sdb1) succeeded (0.0 sec)
Stop successful.
= FILESYSTEM STATUS (bwork) =
TYPE # STATUS NODES
---- - ------ -----
MGT 1 online node115
[root@node115 ~] # echo $?
0
[root@node115 ~] # df
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/sdb1 54420 6924 44688 14% /mnt/bwork/mgt
/dev/sdb2 365364336 36043708 304958460 11% /mnt/bwork/mdt/0
Reported by: thiell
degremont commented
- Version: --> 1.3.1
Original comment by: thiell
degremont commented
- status: new --> closed
- assigned_to: Aurélien Degrémont
- Resolution: --> fixed
Original comment by: degremont
degremont commented
Fixed in [f4e89b]
This could be improved later either unmounting using the mount path and/or checking expected state again at ev_close in case of success.
Original comment by: degremont