Mirantis/sriov-scheduler

Name: sriov-scheduler

Owner: Mirantis Inc.

Description: null

Created: 2017-11-02 09:11:18.0

Updated: 2017-11-02 09:51:57.0

Pushed: 2017-11-08 08:55:17.0

Homepage: null

Size: 47

Language: Go

GitHub Committers

User	Most Recent Commit	# Commits

Other Committers

User	Email	Most Recent Commit	# Commits

README

Sriov scheduler extension

Demo

Get started

This application does 2 things:

• discovers total number of sriov vfs on every node in the cluster
• prevents scheduler from binding pods on a nodes that doesn't have vfs left

In order to deploy applicaton you need to follow next instructions. Run a disvery tool on every node in the cluster:

ctl create -f tools/discovery.yaml

It will deploy daemonset with golang script that will read totalvfs number on each node and report them to kubernetes. Reported information will be available on a nodes:

atus:
allocatable:
  cpu: "8"
  memory: 32766568Ki
  pods: "110"
  totalvfs: "1"

Next deploy scheduler extension itself:

ctl create -f tools/extender.yaml

It will create deployment with http server and a service for it.

Important to note that all pods without requested sriov network will be ignored, in future it will be easy to add any other selection algorithm:

: Pod
data:
notations:
networks: sriov

And as a last step we need to change kubernetes scheduler configuration. On my environment kubernetes scheduler is self-hosted and I will be using configmap as a policy configuration source.

Create configmap “scheduler-policy” in a kube-system namespace:

ctl create configmap scheduler-policy -n kube-system --from-file=policy.cfg=tools/scheduler.json

And add policy-configmap option to kubernetes scheduler:

:
ntainers:
command:
- /hyperkube
- scheduler
- --address=127.0.0.1
- --leader-elect=true
- --kubeconfig=/etc/kubernetes/scheduler.conf
- --policy-configmap
- scheduler-policy