-
Bug
-
Resolution: Unresolved
-
Undefined
-
None
-
4.15.0
-
None
-
Important
-
None
-
False
-
Description of problem:
A PV was configured with wwid as below:
# oc get pv pv0001 -o yaml |yq '.spec.fc'
wwids:
- 0QEMU_QEMU_HARDDISK_drive-scsi0-0-0-1
Multipath was configured on the node and the disk already had a partition:
# multipath -ll mpathb mpathb (0QEMU_QEMU_HARDDISK_drive-scsi0-0-0-1) dm-0 QEMU,QEMU HARDDISK size=100G features='0' hwhandler='0' wp=rw `-+- policy='service-time 0' prio=1 status=active `- 0:0:0:1 sdb 8:16 active ready running
Device Boot Start End Sectors Size Id Type /dev/disk/by-id/dm-name-mpathb1 2048 2099199 2097152 1G 83 Linux
Then a PVC was created and attached to the pod:
# oc get pod virt-launcher-rhel7-crimson-manatee-46-8r4lb -o yaml |yq '.spec.containers[0].volumeDevices,.spec.volumes[7]' - devicePath: /dev/disk-sapphire-reindeer-40 name: disk-sapphire-reindeer-40 name: disk-sapphire-reindeer-40 persistentVolumeClaim: claimName: pv-new
But the pod can only see 1 GiB:
sh-5.1$ blockdev --getsize64 /dev/disk-sapphire-reindeer-40 1073741824
This is because the partition of the disk was mapped to the pod instead of full disk:
crictl inspect 6d23a623beb6e |grep -A 4 -B 4 "/dev/disk-sapphire-reindeer-40" "uid": 0, "gid": 36 }, { "path": "/dev/disk-sapphire-reindeer-40", "type": "b", "major": 253, "minor": 1, "uid": 0, # lsblk |grep -B 1 "253:1" └─mpathb 253:0 0 100G 0 mpath └─mpathb1 253:1 0 1G 0 part
It looks to be problem in how findDiskWWIDs find the multipath disk:
- It fist looks at /dev/disk/by-id/ and look for scsi-wwid and find the symlink to get the SCSI device name.
- Then it looks at /sys/block/dm-*/slaves/<dev-name> and match it with the found device name in step 1 to find the multipath device name.
However, here /dev/disk/by-id/scsi-wwid will be already pointing to multipath device name.
ls -l /dev/disk/by-id/scsi-0QEMU_QEMU_HARDDISK_drive-scsi0-0-0-1 lrwxrwxrwx. 1 root root 10 Sep 24 06:41 /dev/disk/by-id/scsi-0QEMU_QEMU_HARDDISK_drive-scsi0-0-0-1 -> ../../dm-0
So when it tries to find this device in /sys/block/dm-/slaves/ , it match with the partition:
ls -l /sys/block/dm-1/slaves/ total 0 lrwxrwxrwx. 1 root root 0 Sep 24 06:16 dm-0 -> ../../dm-0
Also, below log shows same:
Sep 24 06:42:32 openshift-worker-cygnus-0 kubenswrapper[3575]: I0924 06:42:32.404613 3575 fc_util.go:106] fc: find disk: /dev/dm-0, dm: /dev/dm-1
Version-Release number of selected component (if applicable):
4.16.8 (however I think it is reproducible on previous versions)
How reproducible:
100%
Steps to Reproduce:
1. Enable multipath on the node.
2. Map a disk which is having partition to the pod using wwid.
3. Pod will see partition instead of whole disk.
Actual results:
FC PV configured with WWID is mapping the partition to the pod instead of LUN
Expected results:
Additional info:
- is related to
-
-
- ON_QA
-
