Skip to content

OpenNebula @ Kubernetes - Deploy OpenNebula on Kubernetes

License

Notifications You must be signed in to change notification settings

ethz-hpc/k8s-OpenNebula

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

46 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Opennebula

opennebula @ kuberenetes

Requierments

  • Helm
  • Helmfile (optional)
  • Kuberenetes +1.10 (tested on 1.16)

Recommended

  • NFS for the datasotarages (system + images)

Installation

cd Chart && helmfile sync

helmfile will install the local chart in opennebula namespace.

with helm

helm install -n opennebula --namespace opennebula chart/

Auto ssh keys

SSH Keys could be automatically created on helm install.

Enable (on values.yaml) auto_ssh to create ssh keys secrets before install the chart. (templates/auto-) If you disable this you can run ./createSshkeys.sh before creating the chart or create them manually. (Note, by default this will push to 'opennebula' namespace)

Auto serveradmin

Because serveradmin user is created by Onedeamon (opennebula core process) we need to create a secret in runtime from that. Enable auto_serveradmin_secret to create the secret from the oned container.

Datastores

Enable in values datasteres to create a PVC. For now, its needed to create the datastores in the UI or in the cli matching the id's. Dont create one before so the ids starts on 100, 101. http://docs.opennebula.org/5.8/deployment/open_cloud_storage_setup/fs_ds.html#shared-qcow2-transfer-modes

>cat ds.conf
NAME   = nfs_images
DS_MAD = fs
TM_MAD = shared
>onedatastore create ds.conf

We are working on OneProvision.

Add Opennebula workers.

There is a deamonset that will ran the worker pod on the annotated nodes using the node-selector feature. https://kubernetes.io/docs/concepts/configuration/assign-pod-node/

In the values files check:

annotate your nodes to join the oppennebula cluster,

if Nil {} this will be deployed in all possible nodes.

    nodeSelector:
      opennebula-node: ""

The worker pod will run the Opennebula node that creates the VM's. This allow us to specify in witch node we want to run this worker pod. In case you are running this on VM's: be aware that nested virtualization could be tricky. So you can have VM's for the control plane and bare metal machines for the workers. In this case anotate the bare-metal nodes to run the worker pod there.

You should see the label something like this:

kubectl get node some-node-here -o yaml
apiVersion: v1
kind: Node
...
  labels:
    opennebula-node: ""

Secure Opennebula.

Its tested to work with ssl termination at traefik ingress with cert manager. For installing cert-manager with helmfile: https://docs.cert-manager.io/en/latest/getting-started/install/kubernetes.html#alternative-installation-methods

How to Upload an image if you dont have internet access :

For now, until we implement something better: scp your image to where oned StatefulSet is running. To get the node where it is running:

kubectl get pods -n opennebula -l app=opennebula-oned -o wide

Use findmnt to find the path where the "data" emptyDir is in that node and put your image there. This data volume is mounted on oned container on /var/tmp, so now you can create a new image from host path /var/temp/tourimage.qcow2

Tested on qcow2 images.

Uninstall Opennebula

cd Chart && helmfile destroy

to delete the secrets created by this chart (not tracked by Helm)

./Chart/deleteall.ch

or

kubectl delete ns opennebula

Secrets

If you enabled the auto_ssh or auto_serveradmin_secret, this resources are not being tracked by helm, so if you delete with helm delete the secret will still be there.

Mysql PVC:

By default in the values the mysql pvc have the annotation keep, to not delete it if we uninstall the chart.

you can run ./deleteall.sh to make shure nothing is there. You can also remove the namespace, named 'openbnebula' in the helmfile.

Possible improvements:

  • Appmarket
  • gate

Writing OneOperator to:

  • When new node is running (now k8s node with the desire label) add it into the Opennebula cluster.
  • If a node is not available, disable it in Opennebula cluster to stop scheduling things there.
  • If a node is going down as a prestop hook: drain opennebula node (move Vm's) before it stop: With this we can change the Nodes demonset to updateStrategy: Rollingupdate (now is OnDelete)

Images

available at

Authors

ETH Zurich, High performance computing team.

Thans to kvaps for sharing.

About

OpenNebula @ Kubernetes - Deploy OpenNebula on Kubernetes

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published