Support hardware-aware affinity for pods on k8s cluster

Registered by Yoshito Ito

In case of deploying a Container Network Function on the VNF of a Kubernetes
cluster with the blueprint "CNF Support with ETSI NFV specifications" [1], the
Pods may be scheduled on the same physical Compute server while they are
labeled with anti-affinity rules [2]. The anti-affinity Rule can deploy the
Pods on different worker nodes, but the worker nodes may be in the same
server.

In this spec, we propose a hardware-aware affinity for Pods. This is achieved
by labeling the worker nodes with the identifier of deployed Compute servers.
In Tacker, MgmtDriver can handle the installation of Kubernetes cluster as a
VNF instance, and the `instantiate_end` script can help the process to make
and assign the labels.

[1] https://blueprints.launchpad.net/tacker/+spec/cnf-support-with-etsi-nfv-specs
[2] https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/

Blueprint information

Status:
Started
Approver:
None
Priority:
Undefined
Drafter:
Yoshito Ito
Direction:
Needs approval
Assignee:
Yoshito Ito
Definition:
New
Series goal:
Proposed for wallaby
Implementation:
Slow progress
Milestone target:
milestone icon wallaby-3
Started by
Yoshito Ito

Related branches

Sprints

Whiteboard

Gerrit topic: https://review.opendev.org/#/q/topic:bp/hardware-aware-pod-affinity

Addressed by: https://review.opendev.org/759774
    Support hardware-aware affinity for pods on k8s cluster

(?)

Work Items

This blueprint contains Public information 
Everyone can see this information.

Subscribers

No subscribers.