[nova][ptg] Expose auto converge and post copy
spec: https://review.opendev.org/#/c/687199 There was multiple discussion during the PTG around this. I think at the end mdbooth and yawang found a possible solution that they liked and sounded OK to me too. Unfortunately the etherpad was not updated and my memory is bad enough that I cannot gather what was the exact proposal. yawang: could you please post a short summary here and / or simply update the spec with the discussed solution? Cheers, gibi
Hi: * The original method(expose the auto converge/post copy in image properties/flavor extra specs) exposed features of hypervisor layer directly, and it affects scheduling. Therefore, it's not appropriate. * The new method is add a new parameter: "no-performance-impact". If the parameter set during live migrate, the libvirt driver will disable the auto converge and post copy functions. User can tag their instances as "no performance impact" in instance metadata or somewhere else, operator can check the tag to decide whether add the parameter before live migrate. I will write a new spec to describe these in detail :) Best Regards On 2019/11/11, 12:15 AM, "Balázs Gibizer" <openstack-discuss-bounces+wang.ya=99cloud.net@lists.openstack.org on behalf of balazs.gibizer@est.tech> wrote: spec: https://review.opendev.org/#/c/687199 There was multiple discussion during the PTG around this. I think at the end mdbooth and yawang found a possible solution that they liked and sounded OK to me too. Unfortunately the etherpad was not updated and my memory is bad enough that I cannot gather what was the exact proposal. yawang: could you please post a short summary here and / or simply update the spec with the discussed solution? Cheers, gibi
Hi: Here is the spec [1]_ Because the exist spec [2]_ has gap with the agreement, so I rewrote a new spec. .. [1]: https://review.opendev.org/#/c/693655/ .. [2]: https://review.opendev.org/#/c/687199/ Best Regards On 2019/11/11, 10:43 AM, "wang.ya" <openstack-discuss-bounces+wang.ya=99cloud.net@lists.openstack.org on behalf of wang.ya@99cloud.net> wrote: Hi: * The original method(expose the auto converge/post copy in image properties/flavor extra specs) exposed features of hypervisor layer directly, and it affects scheduling. Therefore, it's not appropriate. * The new method is add a new parameter: "no-performance-impact". If the parameter set during live migrate, the libvirt driver will disable the auto converge and post copy functions. User can tag their instances as "no performance impact" in instance metadata or somewhere else, operator can check the tag to decide whether add the parameter before live migrate. I will write a new spec to describe these in detail :) Best Regards On 2019/11/11, 12:15 AM, "Balázs Gibizer" <openstack-discuss-bounces+wang.ya=99cloud.net@lists.openstack.org on behalf of balazs.gibizer@est.tech> wrote: spec: https://review.opendev.org/#/c/687199 There was multiple discussion during the PTG around this. I think at the end mdbooth and yawang found a possible solution that they liked and sounded OK to me too. Unfortunately the etherpad was not updated and my memory is bad enough that I cannot gather what was the exact proposal. yawang: could you please post a short summary here and / or simply update the spec with the discussed solution? Cheers, gibi
On Mon, Nov 11, 2019 at 15:45, wang.ya <wang.ya@99cloud.net> wrote:
Hi:
Here is the spec [1]_ Because the exist spec [2]_ has gap with the agreement, so I rewrote a new spec.
.. [1]: https://review.opendev.org/#/c/693655/ .. [2]: https://review.opendev.org/#/c/687199/
Could you please abandon one of the specs this is no confuses me which solution you want to push forward. Cheers, gibi
Best Regards
On Mon, Nov 11, 2019 at 15:45, wang.ya <wang.ya@99cloud.net> wrote:
Hi:
Here is the spec [1]_ Because the exist spec [2]_ has gap with the agreement, so I rewrote a new spec.
.. [1]: https://review.opendev.org/#/c/693655/ .. [2]: https://review.opendev.org/#/c/687199/ Could you please abandon one of the specs this is no confuses me which solution you want to push forward.
https://review.opendev.org/#/c/687199/ has been abandoned Please discuss in this spec: https://review.opendev.org/#/c/693655/;-)
participants (3)
-
Balázs Gibizer
-
wang.ya
-
wangya