tests.md

Single Network

This is to test the basic functionality of ovn-kubernetes when only the default networks exists.

Single Node

To install a single node running both master and minion for k8s as well as ovn-kubernetes use:

./install-ovn-kubernetes.sh

And then run:

./run-ovn-kubernetes.sh

The ovn-kubernetes log will be at /var/log/openvswitch/ovnkube.log To create the pods, use:

kubectl create -f cirros-pod.yaml
kubectl create -f another-cirros-pod.yaml

To see that the pods are running and have IPs in the correct range, use:

kubectl get pods -o wide

To test connectivity we want to test ping between the pods:

CIRROS_IP=`kubectl exec cirros ip addr | grep "10.244" | cut -f1 -d/ | awk '{print $2}'`
kubectl exec another-cirros ping $CIRROS_IP

Multiple Nodes

To install multiple nodes running one master for both k8s and ovn-kubernetes and minions use, on the master:

./install-ovn-kubernetes-master.sh
./run-ovn-kubernetes-master.sh

And on each minion (note that some interaction will be needed):

./install-ovn-kubernetes-minion.sh <master ip>
./run-ovn-kubernetes-minion.sh

And then run (on master or minion):

kubectl create -f cirros-pod.yaml
kubectl create -f another-cirros-pod.yaml

To see that the pods are running and have IPs in the correct range, use:

kubectl get pods -o wide

To test connectivity we want to test ping between the pods:

CIRROS_IP=`kubectl exec cirros ip addr | grep "10.244" | cut -f1 -d/ | awk '{print $2}'`
kubectl exec another-cirros ping $CIRROS_IP

Different Pods on Different Nodes

TODO

Multiple Network

This is to tests the enhanced functionality of multiple networks with ovn-kubernetes, as enhanced by this fork

Single Node

To install a single node running both master and minion for k8s as well as ovn-kubernetes use:

./install-ovn-kubernetes-multinet.sh

Because an additional network is needed, an additional switch needs to be added to the node. This is done by:

ovn-nbctl ls-add an-ovn-switch1
ovn-nbctl ls-add an-ovn-switch2
ovn-nbctl set logical_switch an-ovn-switch1 other-config:subnet=10.0.0.0/24
ovn-nbctl set logical_switch an-ovn-switch2 other-config:subnet=10.10.0.0/24

Now run ovn-kubernetes (similarly to the single net case):

./run-ovn-kubernetes.sh

The ovn-kubernetes log will be at /var/log/openvswitch/ovnkube.log

Now the NetworkAttachmentDefinitions CRDs need to be created:

kubectl create -f ovn-network1.yaml 
kubectl create -f ovn-network2.yaml 

To create the pods, use:

kubectl create -f cirros-pod-multinet.yaml
kubectl create -f another-cirros-pod-multinet.yaml

To test connectivity we want to test ping between the pods:

CIRROS_IP=`kubectl exec cirros-multinet ip addr | grep "10.10" | cut -f1 -d/ | awk '{print $2}'`
kubectl exec another-cirros-multinet ping $CIRROS_IP

Connect to Physical Network

Currently blocked by OVN issue Would work only if static MAC is provided in pod annotation. For example:

annotations:
   switches: a-physnet-switch
   k8s.v1.cni.cncf.io/networks: an-ovn-physnet
   ovn_extra: '{"a-physnet-switch":{"mac_address":"0a:00:00:00:00:01"}}'

For that, it is recommended to use a host with 2 physical interfaces. Assuming the 2nd interface is "eth1", first step is to greate an ovs bridge:

ovs-vsctl add-br breth1

And associate with an ovn physical network:

ovs-vsctl set Open_vSwitch . external-ids:ovn-bridge-mappings=phyNet:breth1

Now create the ovn logical bridge and ports (for physical network):

ovn-nbctl ls-add a-physnet-switch
ovn-nbctl lsp-add a-physnet-switch physport-localnet

Or, alternatively, replace the last line with:

ovn-nbctl lsp-add a-physnet-switch physport-localnet parent 101

To mark that port with VLAN tag 101 (instead of the default one).

ovn-nbctl lsp-set-addresses physport-localnet unknown
ovn-nbctl lsp-set-type physport-localnet localnet
ovn-nbctl lsp-set-options physport-localnet network_name=phyNet

Last, associate the ovs bridge with the physical NIC:

ovs-vsctl add-port breth1 eth1

Now create the NetworkAttachmentDefinition CRD for that new switch:

kubectl create -f ovn-physnet.yaml 

And create a pod that uses it:

kubectl create -f cirros-pod-physnet.yaml

To make sure that the pod is indeed connected to the physical network, use pods with tcpdump, ip and ping installed:

kubectl create -f fedora-pod-physnet.yaml
kubectl create -f another-fedora-pod-physnet.yaml

Make sure that the additional interfaces were added on the pods:

kubectl exec -it fedora-physnet ip link
kubectl exec -it another-fedora-physnet ip link

Then allocate static IPs on these interfaces with their routes:

kubectl exec -it fedora-physnet ip addr add 192.168.50.2 dev net1
kubectl exec -it fedora-physnet ip route add 192.168.50.0/24 dev net1
kubectl exec -it another-fedora-physnet ip addr add 192.168.50.3 dev net1
kubectl exec -it another-fedora-physnet ip route add 192.168.50.0/24 dev net1

Then ping from one pod:

kubectl exec -it fedora-physnet ping 192.168.50.3

And tcpdump on the other:

kubectl exec -it another-fedora-physnet tcpdump -- -n -i net1

Multiple Node

TODO