The multipath device descriptors remove failed when the volume has partition

[weiweigu]

tested environment:
iscsi_use_multipath=True and iscsI volume backed is FUJITSU.
An instance boots form a volume which has a image with partition.

After terminating the instance，I found the multipath device "/dev/mapper/mpathq" deleted faild.
And the sd* devices have been deleted success.

[root@opencos170 /(keystone_admin)]# multipath -l /dev/mapper/mpathq
mpathq (3600000e00d280000002800e2000f0000) dm-2
size=201G features='0' hwhandler='0' wp=rw
[root@opencos170 /(keystone_admin)]#

Can't delete the multipath device to use command "multipath -f /dev/mapper/mpathq".
[root@opencos170 /(keystone_admin)]# multipath -f /dev/mapper/mpathq
Jan 26 10:40:37 | mpathqp3: map in use
Jan 26 10:40:37 | failed to remove multipath map /dev/mapper/mpathq
[root@opencos170 /(keystone_admin)]#

[Sean Dague (sdague)]

is there a generic reproduce on this for something like the lvm driver? Basically is this an issue with OpenStack code or with something below us? In this bug's current state it's not very actionable.

[Launchpad Janitor (janitor)]

[Expired for OpenStack Compute (nova) because there has been no activity for 60 days.]

[Matt Riedemann (mriedem)]

Please reply to Sean's question in comment 1. Also, which version of nova is this? Can it be reproduced in Liberty (which is when the nova libvirt volume modules started using the os-brick library which is shared with Cinder).

[weiweigu (gu-weiwei)]

We use icehouse and kilo version, they have the same problem. Liberty is not used.

[Launchpad Janitor (janitor)]

[Expired for OpenStack Compute (nova) because there has been no activity for 60 days.]

[Pooja Ghumre (pooja-9)]

Reopening this issue since we are able to consistently reproduce it with Nova on Rocky and Cinder on Pike as well. Is there any potential fix in os-brick that can help with this? In our case, this was seen with PureStorage volume backend.

[Pooja Ghumre (pooja-9)]

Should this bug be filed under os-brick if Nova is not the right place to fix this multipath issue?

[MarkMielke (mark-mielke)]

I recently hit this when switching storage drivers - particularly from a single path driver to a multi path driver, and the cause seems to have only been LVM. I had previously specified the prior iSCSI system via /etc/lvm/lvm.conf in global_filters, but I missed added the multipath devices to global_filters. The result was that LVM on the hypervisor was seeing LVM-based guest volumes. After adding the multipath devices to global_filters and cleaning up the mess via dmsetup, the problem seems resolved. So, I don't think this is specifically an os-brick issue unless os-brick can block LVM activation on the host for guest volumes on a case-by-case basis, rather than as a global filter.