-
Notifications
You must be signed in to change notification settings - Fork 147
Spec: SRIOV Forwarder #2072
Comments
@rktidwell @rdimitrov @nickolaev @edwarnicke @davecremins Can you please take a look? It's a rough draft as we discussed on the weekly meeting. Mostly describes current implementation and assumptions, and more importantly lists the challenges/open issues. |
If I get it right, I am a big NO here! I am sorry but I don't see the NSM spirit and concepts applied here, instead, this spec os proposing to tweak and patch a ton of code just to be able to request a simple interface.
The workflow should rather be:
I think we should drop the whole Device Plugin idea and figure out a solution independent from it. Otherwise we will have to bend our principles and ideas and we endup with a Multus variation which is not the point at all with NSM. |
Just a small comment so we're all aware - scheduling isn't the main reason for using device plugin, main reason is that it already provides a ton of features like device discovery with selectors for vendors/drivers/device types, health checking of the links, grouping devices into the resource pools, finally allocation the device resources themselves based on what needs to be allocated, I think you may be missing a good number of benefits. And dropping DP results in a need of re-implementing almost all of this in NSM. |
I assume you are referring to this plugin: If that is the case, can we just use whatever lives here: And replace this part with an NSM forwarder logic: I mean, the SRIOV plugin has already what is needed, that is not the argument here. The argument is how we should interface this functionality. NSM is trying to change the way networking is done. We are inspired by the cloud-native principles, we want to target true cloud-native, declarative APIs. What I see in the device plugin framework (kubelet implementation etc) is more suitable to what CNI and Multus do. That is fine, but we need something different. If this means we should challenge the implementation of the existing components- so be it. Otherwise, we are reimplementing the same model and call it different. |
Copying what was discussed in #nsm-dev on Slack, for tracking the discussion:
|
Agree to make the impact as low as possible (preferably none) and I hope the ongoing refactoring work should make that easier. I'd like to invite everyone interested to comment here: networkservicemesh/api#15 |
This issue has been automatically marked as stale because it has not had activity in 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions. |
Link to the Google Docs document
https://docs.google.com/document/d/1TDCUgJ-qAPtFI3Vd1_xR-HSFQh5GyDVnbbrHXasculM/edit#
References
WIP PR #1925
SRIOV Net DP PR k8snetworkplumbingwg/sriov-network-device-plugin#194
Initial spec #727
The text was updated successfully, but these errors were encountered: