On Tue, May 7, 2024 at 10:26 PM <smooney@redhat.com> wrote:

On Tue, 2024-05-07 at 23:58 +0900, Takashi Kajinami wrote:
>
>
> On 5/7/24 22:51, Nguyễn Hữu Khôi wrote:
> > Hello.
> > I do filter as
> >
> > https://docs.openstack.org/cinder/pike/install/cinder-storage-install-ubuntu.html <
> > https://docs.openstack.org/cinder/pike/install/cinder-storage-install-ubuntu.html>
>
> To make sure that we have no misunderstanding between us, did you configured the filter
> not only in storage node (or probably controller node if you don't have dedicated nodes
> for cinder-volume) BUT ALSO in all compute nodes ?
right this is imporant because iscsi volumes are host mounted on the compute nodes
and if you dod not have the filter configured the host can mount the file systems and that causes
the device to be coniserded busy when disconnecting

tripleo has some docs on this here
https://docs.openstack.org/project-deploy-guide/tripleo-docs/latest/features/lvmfilter.html
but it applies to any intaller.

it was implented in
https://github.com/openstack-archive/tripleo-ansible/tree/stable/zed/tripleo_ansible/roles/tripleo_lvmfilter

triple just invokes lvmconfig to generate a file and then copies it to /etc/lvm/lvm.conf

in general you want to ensure on your compute nodes that lvm filterign is enabled and that you only allow the specific
devices used by the OS not any iscsi devices.

>
> I'm asking this because the current documentation is not specific about this requirement
> but the filter should be configured in all nodes which may attach instance volumes.
>
> >
> > I check and dont see any wrong mapping. I also set skip_kpartx "yes".
>
> So you mean you don't see any lvms from guest instances when you do lvs/vgs/pvs in your
> compute nodes ?
>
> >
> > I still find a solution for it or atleast when resizing/migrating failed, my instance wont be set to ERROR state.
> >
> > This problem happens only with iscsi.
> >
> > On Tue, May 7, 2024, 8:12 PM Takashi Kajinami <kajinamit@oss.nttdata.com <mailto:kajinamit@oss.nttdata.com>> wrote:
> >
> >     The failure reminds me of the "well-known" problem caused by lvm.
> >     I'd suggest that you check
> >
> >      - If instances which fails to be resized/migrated have lvm deployed in their volumes
> >
> >      - If these lvms on instance volumes are detected by host operating system
> >
> >      - If you configure appropriate lvm filter in your compute nodes to avoid host lvm system
> >      from detecting lvms on devices for instance volumes.
> >
> > https://docs.openstack.org/cinder/latest/install/cinder-storage-install-rdo.html#prerequisites <https://docs.openstack.org/cinder/latest/install/cinder-storage-install-rdo.html#prerequisites
> > >
> >
> >
> >     On 5/6/24 23:38, Nguyễn Hữu Khôi wrote:
> >      > Hello.
> >      >
> >      > I am using Openstack Yoga with Dell Powerstore 5000T is backend.
> >      >
> >      > This is my report:
> >      >
> >      > https://bugs.launchpad.net/os-brick/+bug/2063345 <https://bugs.launchpad.net/os-brick/+bug/2063345>
> > <https://bugs.launchpad.net/os-brick/+bug/2063345 <https://bugs.launchpad.net/os-brick/+bug/2063345>>
> >      >
> >      > This bug is also with Victoria and NetApp as backend:
> >      >
> >      > https://bugs.launchpad.net/os-brick/+bug/1992289 <https://bugs.launchpad.net/os-brick/+bug/1992289>
> > <https://bugs.launchpad.net/os-brick/+bug/1992289 <https://bugs.launchpad.net/os-brick/+bug/1992289>>
> >      >
> >      > It looks like with iscsi multipath only. I don't see any reports with Ceph.
> >      >
> >      > I try to create problem but It is random when I do resize instance(instance will move to another compute-cold
> > migrate)
> >      >
> >      > My log tells that
> >      >
> >      > 2022-10-04 10:50:12.238 7 ERROR oslo_messaging.rpc.server Command: multipath -f
> > 3600a09803831486e695d536269665144
> >      > 2022-10-04 10:50:12.238 7 ERROR oslo_messaging.rpc.server Exit code: 1
> >      > 2022-10-04 10:50:12.238 7 ERROR oslo_messaging.rpc.server Stdout: ''
> >      > 2022-10-04 10:50:12.238 7 ERROR oslo_messaging.rpc.server Stderr: 'Oct 04 10:49:32 |
> > 3600a09803831486e695d536269665144: map in use\n
> >      >
> >      >
> >      > Nguyen Huu Khoi
> >      >
> >      >
> >      > On Mon, May 6, 2024 at 9:22 PM Eugen Block <eblock@nde.ag <mailto:eblock@nde.ag>
> > <mailto:eblock@nde.ag <mailto:eblock@nde.ag>>> wrote:
> >      >
> >      > Hi,
> >      >
> >      > I think you'll need to be more specific here. Which openstack version
> >      > are you running? Just recently I have been shuffling around instances
> >      > with live and cold migration (in version Victoria, so quite old), some
> >      > of them did fail but I didn't have to reset any state or
> >      > shelve/unshelve instances. I can't recall if the instances were in
> >      > ERROR state, though. I didn't need to check because after fixing
> >      > whatever broke the migration I just issued another (live-)migrate
> >      > command and then it worked.
> >      > We use only ceph as backend for glance, nova and cinder, which
> >      > backend(s) do you use? Maybe that has an impact as well, I don't know.
> >      >
> >      > Regards,
> >      > Eugen
> >      >
> >      > Zitat von Nguyễn Hữu Khôi <nguyenhuukhoinw@gmail.com <mailto:nguyenhuukhoinw@gmail.com>
> > <mailto:nguyenhuukhoinw@gmail.com <mailto:nguyenhuukhoinw@gmail.com>>>:
> >      >
> >      > > Hello.
> >      > >
> >      > > I would like to know why we set instance-state to ERROR when
> >      > > resizing/migrating failed? I have a problem with resizing/migrating
> >      > > instances. I must reset state then shelve and unshelve instances to make it
> >      > > come back.
> >      > >
> >      > > Thank you for your time.
> >      > >
> >      > > Nguyen Huu Khoi
> >      >
> >      >
> >      >
> >
>