Available device bandwidth report by L2 agent

There's very little to this RFE that would allow us to make an informed decision. Please consider providing more details.

docs.openstack.org/developer/neutron/policies/blueprints.html#neutron-request-for-feature-enhancements

Bad paste:

http://docs.openstack.org/developer/neutron/policies/blueprints.html#rfe-submission-guidelines

I am not sure putting the data into state reports is the way to go. My limited understanding of nova scheduler makes me think that compute nodes report their data significant for scheduling decisions to nova scheduler directly, omitting controller (at least neutron controller). Of course, being able to push neutron bandwidth data (and make decision based on it) into nova scheduler would require some pluggable framework in nova. Such framework was discussed multiple times with those involved in nova scheduler on the latest summit and in other venues.

I thought that Miguel (ajo) was going to propose a nova spec to drive the pluggable scheduler effort. Not sure what's the status there. Adding Miguel to subscribed.

Hi Ihar, the nova scheduler spec you were asking about is this one: https://review.openstack.org/#/c/252395/

It's very preliminary, and probably inconsistent in many forms yet, but it's up for discussion, I have to re-read jay's resource aggregation spec for the nova scheduler to see if it does (somehow) fit our scheduling purposes.

What Irena is proposing here (I asked her to help with proposing this RFE), is a first step, for a later information feed back to nova.

May be it's too early for this, and we will need to see who feeds the information exactly (compute nodes or neutron-server).

It could be interesting, not only to feed nova, but to provide an API to ask neutron-server, "which hosts can I use to plug this port meeting all it's characteristics" (note: this is not desired by nova because they don't want to do an external API call from the scheduler) but other projects integrating to neutron could benefit from it. From that point of view, centralising the information is a good idea.

I suggest we wait until we are clear who will feed the scheduler with the data. There is no reason to push it into server if computes will send it to the scheduler.

[Expired for neutron because there has been no activity for 60 days.]

+1 this physical capability should be known by agent for better service.
and it would be nice to have a validation on user input as well if that's really really high
than highest available physical capability of any host.

I think that this will be now done with placement API, and is described in specs: https://review.openstack.org/#/c/508149/