The openstack-helm project can be used to install a set of Kubernetes nodes as OpenStack compute nodes, with the OpenStack control services (nova, neutron, keystone, glance, etc.) running as containers on Kubernetes. This is an easy way to bring up an OpenStack cluster that can be controlled by loading the Base OpenStack chart.
Instructions for installing openstack-helm on a single node or a multi-node cluster can be found at https://docs.openstack.org/openstack-helm/latest/index.html.
The following describes steps for installing openstack-helm, including how to customize the documented install procedure with specializations for CORD. Specifically, CORD uses the VTN ONOS app to control Open vSwitch on the compute nodes and configure virtual networks between VMs on the OpenStack cluster. Neutron must be configured to pass control to ONOS rather than using openvswitch-agent to manage OvS.
After the install process is complete, you won't yet have a fully-working OpenStack system; you will need to install the base-openstack chart first.
For convenience, a script to install Kubernetes, Helm, and openstack-helm on a single Ubuntu 16.04 node is provided in the automation-tools repository. This script also customizes the install as described below.
git clone https://gerrit.opencord.org/automation-tools automation-tools/openstack-helm/openstack-helm-dev-setup.sh
If you run this script you can skip the instructions on the rest of this page.
To enable the VTN app to control Open vSwitch on the compute nodes, it is necessary to customize the openstack-helm installation. The customization occurs through specifiying values.yaml files to use when installing the Helm charts.
The openstack-helm installation process designates one node as the master node; the Helm commands are run on this node. The following values files should be created on the master node prior to installing the openstack-helm charts.
cat <<EOF > /tmp/glance-cord.yaml --- network: api: ingress: annotations: nginx.ingress.kubernetes.io/proxy-body-size: "0" EOF export OSH_EXTRA_HELM_ARGS_GLANCE="-f /tmp/glance-cord.yaml"
cat <<EOF > /tmp/libvirt-cord.yaml --- network: backend: [] EOF export OSH_EXTRA_HELM_ARGS_LIBVIRT="-f /tmp/libvirt-cord.yaml"
cat <<EOF > /tmp/nova-cord.yaml --- labels: api_metadata: node_selector_key: openstack-helm-node-class node_selector_value: primary network: backend: [] pod: replicas: api_metadata: 1 placement: 1 osapi: 1 conductor: 1 consoleauth: 1 scheduler: 1 novncproxy: 1 EOF export OSH_EXTRA_HELM_ARGS_NOVA="-f /tmp/nova-cord.yaml"
cat <<EOF > /tmp/neutron-cord.yaml --- images: tags: neutron_server: xosproject/neutron-onos:newton manifests: daemonset_dhcp_agent: false daemonset_l3_agent: false daemonset_lb_agent: false daemonset_metadata_agent: false daemonset_ovs_agent: false daemonset_sriov_agent: false network: backend: [] interface: tunnel: "eth0" pod: replicas: server: 1 conf: plugins: ml2_conf: ml2: type_drivers: vxlan tenant_network_types: vxlan mechanism_drivers: onos_ml2 ml2_type_vxlan: vni_ranges: 1001:2000 onos: url_path: http://onos-ui.default.svc.cluster.local:8181/onos/cordvtn username: onos password: rocks EOF export OSH_EXTRA_HELM_ARGS_NEUTRON="-f /tmp/neutron-cord.yaml"
It is also necessary to make a small change to openstack-helm's openvswitch chart: the /usr/sbin/ovsdb-server must be executed with the --remote=ptcp:6641 option to listen for the connection from VTN. After the openstack-helm repository is checked out during the install process, run the following command:
cd openstack-helm/openvswitch/templates/bin sed -i 's/--remote=db:Open_vSwitch,Open_vSwitch,manager_options/--remote=db:Open_vSwitch,Open_vSwitch,manager_options --remote=ptcp:6641/' _openvswitch-db-server.sh.tpl
Please see the openstack-helm documentation for instructions on how to install openstack-helm on a single node (for development and testing) or a multi-node cluster.
The install process is flexible and fairly modular; see the links above for more information. At a high level, it involves running scripts to:
openstack-helm install overwrites /etc/resolv.conf on the compute hosts and points the upstream nameservers to Google DNS. If a local upstream is required, see this note.)values.yaml files based on the environment and install Helm charts using these files