[openstack-dev] [qa] New upgrade test tool proposal.

Sean Dague sean at dague.net
Tue Jun 27 19:35:11 UTC 2017

On 06/27/2017 03:19 PM, Dean Troyer wrote:
> On Tue, Jun 27, 2017 at 1:37 PM, Jay Pipes <jaypipes at gmail.com> wrote:
>> Hi Castulo, sorry for the delayed response on this. Has your team moved
>> forward on any of this?
> IIRC this work was impacted by the OSIC shutdown, I believe it is not
> currently on anyone's radar.
>> What about the Grenade testing framework that uses devstack as its
>> deployment system was not useful or usable for you?
> I can take at least part of the blame in encouraging them to not
> attempt to leverage Grenade directly.  Grenade needs to be replaced as
> it has far out-lived its expected life[0].
> Grenade was built to do static in-place upgrades and the fact that it
> has been pushed as far as it has is a happy surprise to me.  However,
> it is fundamentally limited in its abilities as a test orchestrator,
> implementing robust multi-node capabilities and the granularity that
> is required to properly do upgrade testing really needs a reboot.  In
> a well-funded world that would include replacing DevStack too, which
> while nice is not strictly necessary to achieve the testing goals they
> had.
> The thing that Grenade and DevStack have going for them besides
> inertia is that they are not otherwise tied to any deployment
> strategy.  Starting over from scratch really is not an option at this
> point, something existing really does need to be leveraged even though
> it may hurt some feelings along the way for the project(s) not chosen.
> dt
> [0] Seriously, I never expected Grenade (or DevStack for that matter)
> to have survived this long, but they have mostly because they were/are
> just barely good enough that nobody wants to fund replacing them.

Well, we also lengthened their life and usefulness by adding an external
plugin interface, that let people leverage what we had without having to
wait on review queues.

But, I also agree that grenade is largely pushed to its limits. That
also includes the fact that I'm more or less the only "active" reviewer
at this point. The only way this is even vaguely tenable is by the
complexity being capped so it's straight forward enough to think through
implications of patches.

The complicated upgrade orchestration including multiple things
executing at the same time, breaks that. It's cool if someone wants to
do it, however it definitely needs more dedicated folks on the review
side to be successful.


Sean Dague

More information about the OpenStack-dev mailing list