[Cinder][driver][ScaleIO]

Martin Chlumsky martin.chlumsky at gmail.com
Wed Feb 6 15:19:53 UTC 2019


Hi Yury,

Thank you for the clarification.
So if we get volumes that are still mapped to hypervisors after deleting
the attached instances with sio_unmap_volume_before_deletion set to False,
there's a good chance it's a bug? I will open a bug report in this case.

Cheers,

Martin

On Wed, Feb 6, 2019 at 9:35 AM Kulazhenkov, Yury <Yury.Kulazhenkov at dell.com>
wrote:

> Hi Martin,
>
> Martin wrote:
> > It seems you would always
> > want to unmap the volume from the hypervisor before deleting it.
> If you remove or shelve instance from hypervisor host, then nova will
> trigger ScaleIO to unmap volume from that host.
> No issues should happen during deletion at this point, because volume is
> already unmapped(unmounted).
> No need to change sio_unmap_volume_before_deletion default value here.
>
> Martin wrote:
> > What is the reasoning behind this option?
> Setting sio_unmap_volume_before_deletion option to  True means that cinder
> driver will force unmount volume from ALL ScaleIO client nodes (not only
> Openstack nodes) during volume deletion.
> Enabling this option can be useful if you periodically detect compute
> nodes with unmanaged ScaleIO volume mappings(volume mappings that not
> managed by Openstack) in your environment.  You can get such unmanaged
> mappings in some cases, for example if there was hypervisor node power
> failure. If during that power failure instances with mapped volumes were
> moved to another host, than unmanaged mappings may appear on failed node
> after its recovery.
>
> Martin wrote:
> >Why would we ever set this
> > to False and why is it False by default?
> Force unmounting volumes from ALL ScaleIO clients is additional overhead.
> It doesn't required in most environments.
>
>
> Best regards,
> Yury
>
> -----Original Message-----
> From: Arkady.Kanevsky at dell.com <Arkady.Kanevsky at dell.com>
> Sent: Wednesday, February 6, 2019 7:24 AM
> To: jsbryant at electronicjungle.net; openstack-discuss at lists.openstack.org;
> Walsh, Helen; Belogrudov, Vladislav
> Subject: RE: [Cinder][driver][ScaleIO]
>
> Adding Vlad who is the right person for ScaleIO driver.
>
> -----Original Message-----
> From: Jay Bryant <jungleboyj at gmail.com>
> Sent: Tuesday, February 5, 2019 5:30 PM
> To: openstack-discuss at lists.openstack.org; Walsh, Helen
> Subject: Re: [Cinder][driver][ScaleIO]
>
> Adding Helen Walsh to this as she may be able to provide insight.
>
> Jay
>
> On 2/5/2019 12:16 PM, Martin Chlumsky wrote:
> > Hello,
> >
> > We are using EMC ScaleIO as our backend to cinder.
> > When we delete VMs that have attached volumes and then try deleting
> > said volumes, the volumes will sometimes end in state error_deleting.
> > The state is reached because for some reason the volumes are still
> > mapped (in the ScaleIO sense of the word) to the hypervisor despite
> > the VM being deleted.
> > We fixed the issue by setting the following option to True in
> cinder.conf:
> >
> > # Unmap volume before deletion. (boolean value)
> > sio_unmap_volume_before_deletion=False
> >
> >
> > What is the reasoning behind this option? Why would we ever set this
> > to False and why is it False by default? It seems you would always
> > want to unmap the volume from the hypervisor before deleting it.
> >
> > Thank you,
> >
> > Martin
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20190206/a605dac4/attachment.html>


More information about the openstack-discuss mailing list