Deploying ONOS classic with HELM
Like the µONOS services, ONOS classic (Java-Karaf based SDN controller) can also be deployed to a Kubernetes cluster.
Local Deployment Setup with Kind
See deploy with helm for a general descriptions for Kind and Helm installation.
Bare metal deployment
For High Availability features the deployment should have at least 3 nodes in the cluster. Because "PodAffinity" is enabled by default - the onos-classic service will not start without 3 nodes with the default settings.
Follow the same Prerequisites as in the µONOS deployment (except the CORD repo is not necessary).
onos-classic can be deployed alongside the µONOS umbrella chart or any combination of individual charts.
Deploy the full HA scenario
This requires at least 3 nodes in the K8S cluster.
helm -n micro-onos install onos-classic onosproject/onos-classic
On kind you can use the following configuration to create a 3 node worker k8s deployment.
kind: Cluster apiVersion: kind.sigs.k8s.io/v1alpha3 nodes: - role: control-plane - role: worker - role: worker - role: worker
After saving this configuration in a file, e.g.
cluster.cfg, you can create the cluster with:
kind create cluster --name onos-classic --config cluster.cfg
Disable Atmoix persistent volumes for clean state after restart
Atomix comes with the capability to save data to persistent volumes to preserve it across re-starts of the containers. For easiness of development lifecycle and to achieve a clean slate every time such feature needs to be disabled. It's disabled by passing
--set atomix.persistence.enabled=false So the resulting command will be
helm -n micro-onos install onos-classic onosproject/onos-classic --set atomix.persistence.enabled=false
You can verify that no persistent volumes are present with
kubectl -n micro-onos get pvc
Deploy on a single node Kind installation
This is the simplest possible deployment with a local Atomix installation and a single replica.
helm -n micro-onos install onos-classic onosproject/onos-classic --set atomix.replicas=0 --set replicas=1
Kind will pull several images from Docker in to its own repo before it starts the cluster. This can be time consuming over a limited connection. It may help performance by doing
docker pullon each of the images listed below just once in to your local registry, and then after Kind starts up loading these in to Kind with:
kind load docker-image ubuntu:16.04 kind load docker-image onosproject/onos:2.2.2 kind load docker-image tutum/dnsutils:latest kind load docker-image atomix/atomix:3.1.0
Accessing the ONOS CLI
The CLI can be accessed through SSH (or if it is available the classic
On a bare metal cluster:
ssh -p 8101 onos@<cluster ip>
the password as usual is
rocks. Access can also be got by
Alternatively using the
onos script from
onos onos@<cluster ip>
On Kind, because NodePorts are not exposed, it is necessary to set up a
port-forward to the
onos-classic pod in the cluster, before accessing the CLI.
kubectl -n micro-onos port-forward $(kubectl -n micro-onos get pods -l app=onos-classic-onos-classic -o name) 8101
The CLI is then available at
ssh -p 8101 onos@localhost
Accessing the ONOS GUI
The GUI can be accessed at port 8181 of the cluster like http://
With Kind it is necessary to set up port-forwarding to port 8181 and the GUI will be available at http://localhost:8181/onos/ui
Controlling the default set of applications
By default only the following apps are installed:
* 28 org.onosproject.optical-model 2.2.2 Optical Network Model * 29 org.onosproject.openflow-base 2.2.2 OpenFlow Base Provider * 34 org.onosproject.drivers 2.2.2 Default Drivers * 83 org.onosproject.gui2 2.2.2 ONOS GUI2
To load additional apps on startup, add them like:
helm -n micro-onos install onos-classic onosproject/onos-classic \ --set atomix.replicas=0 --set replicas=1 \ --set apps=org.onosproject.drivers.bmv2 \ --set apps=org.onosproject.lldpprovider \ --set apps=org.onosproject.hostprovider
i.e. the set used in the NG-SDN Tutorial