[cinder][ops] restricting the Ceph version used with Cinder

Gorka Eguileor geguileo at redhat.com
Fri Nov 13 09:23:27 UTC 2020

On 13/11, Tobias Urdin wrote:
> Hello Brian,
> Thanks for the information! Really appreciate that it’s communicated properly.
> I have a question, I think it’s a little harsh to constraint deployments needing the same release for server and client when The Ceph project itself guarantees backward compatibility. What is the reasoning behind this and will Cinder upstream ensure that upgrade paths between Ceph releases are not broken?
> Best regards


I believe there are multiple reasons to require alignment:

- Without it a system could be using a Luminous client without support
  for a Mimic feature that Cinder wants to use.

- If a really old client has a known bug that won't be fixed, because
  it's no longer supported, we would have to include a workaround in
  Cinder for them all, whereas now we would only have to do it for a
  limited number of versions.

- Backward compatibility at the CLI level doesn't necessarily mean
  backward compatibility at the library level.

- As far as I know, the CI uses aligned versions, so that's what we can
  be sure works.

I'm sure there are other, and probably better, reasons, but in any case,
this reduces the possible combinations of clients-servers that need to
be supported.

As far as the Ceph upgrade paths, I believe upstream Cinder will stay
out of it and let it be a Ceph matter.


> > On 13 Nov 2020, at 01:00, Brian Rosmaita <rosmaita.fossdev at gmail.com> wrote:
> >
> > The Cinder project plans to take advantage of recent developments in Ceph to make some improvements to the RBD driver.  To do that, we propose restricting the Cinder-supported releases of Ceph to the active stable releases plus the two prior releases.
> >
> > This will make Mimic the minimum supported version of Ceph for the Wallaby and X releases of Cinder.
> >
> > This is explained in a bit more detail in a patch to the RBD driver documentation:
> >    https://review.opendev.org/#/c/762592/
> >
> > If you have concerns about this proposal, please leave comments on the review.
> >
> > Thank you,
> > brian
> >

More information about the openstack-discuss mailing list