swift-drive-audit does not action some errors when using disk partitions
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Object Storage (swift) |
New
|
Undecided
|
Unassigned |
Bug Description
I don't see swift-drive-audit unmounted drives after some pretty basic error messages, for example:
# grep sdh /etc/fstab
/dev/sdh1 /srv/node/sdh1 xfs noatime,
root@SM-
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298588] end_request: critical target error, dev sdh, sector 5326104280
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298550] sd 4:0:6:0: [sdh] Unhandled sense code
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298554] sd 4:0:6:0: [sdh] Result: hostbyte=invalid driverbyte=
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298558] sd 4:0:6:0: [sdh] Sense Key : Medium Error [current] [descriptor]
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298575] sd 4:0:6:0: [sdh] Add. Sense: Unrecovered read error
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298579] sd 4:0:6:0: [sdh] CDB: Read(16): 88 00 00 00 00 01 3d 75 e6 d8 00 00 00 08 00 00
Jan 6 18:38:12 SM-X8DTI36B-S-3-xxx kernel: [369063.298588] end_request: critical target error, dev sdh, sector 5326104280
Notice I am using a GPT partition (eg. sdh1 takes up the whole disk sdh)
The error message includes sdh and the kernel_device the script searches for is therefore sdh. It does not match sdh1 in /proc/mounts and therefore does not umount the disk
I understand that using XFS without disks partitions (e.g mkfs.xfs on the underlying /dev/sdxx disk and NOT the partition) is recommended, but I did not know that when I deployed the two clusters that I run and I don't believe it would be possible to move to the underling disks without wiping the disk