Le 11/05/2016 05:12, Jin, Yuntong a écrit :
Hi everyone,
Currently nova exposes all the host CPU instruction set extensions
available
on the compute node in the host state, and there is a scheduler filter
`ComputeCapabilitiesFilter` which looks at these.
But the limits on this is:
CPU instruction set in ComputeCapabilitiesFilter should be guest's
view instead of host's.
Admin may use specific set of CPU instruction to deploy instance to make
it migratable in a heterogeneous cloud.
This is actually by design in nova as nova is using baselineCPU
andallowed to pass/config guest CPU instruction feature for instance.
Shall we add a string “guest_features” in ``ComputeNode`` object as
``ComputeNode:cpu_info:guest_features``
And let ComputeCapabilitiesFilter use guest_features instead of host
features here?
Is this a real gap ? and the above easy fix is the right way ?
FWIW, we had a discussion during the Design Summit on the scheduler Nova
design session about host capabilities and what we call "qualitative
resources" [1]
A first step for helping our users to discover the CPU capabilities is
to provide a Nova abstraction between all our related hypervisor driver
features and you can comment on a proposal [2]
HTH,
-Sylvain
[1] https://etherpad.openstack.org/p/newton-nova-scheduler
[2] https://review.openstack.org/#/c/309762/
Thanks
-yuntongjin
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: [email protected]?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: [email protected]?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
