OpenStack Lab Network - Host Network Interfaces

Configuring network interfaces on OpenStack hosts

Posted by Michael Wadman on January 26, 2019

Overview

This post is the eighth in a series that plans to document my progress through installing and configuring a small OpenStack Lab.

For other posts in this series, see the overview section of the introduction post.

In the last post, we covered the routing configuration on our Cumulus switches.
In this post, we’ll go over the network interface configuration that our OpenStack hosts require in order to participate in the network that we have set up for them.

Configuration

In terms of networking on our OpenStack hosts, we want to accomplish a few things before we even think about installing and running OpenStack itself.
The first, and what we’ll cover in this post, will be configuring our network interfaces to support our routing protocols and our OpenStack configuration.

At this point, it might be important to remind ourselves that we’re preparing our hosts for an OpenStack deployment via OpenStack-Ansible. Their deployment guide contains tons of useful information when going through a first install.
Also of note for this series of posts is their routed environment example, as the configuration involved and the end state of the hosts is similar to what we want to deploy.

Ansible Playbook

As usual, our starting place is the playbook.
Because we’re no longer working with our Cumulus switches and instead on our OpenStack hosts themselves, let’s separate out any roles that we use into a new playbook. It’ll be called “openstack-hosts.yml”

---
- name: Configuring OpenStack hosts
  hosts: openstack_hosts
  become: true
  gather_facts: true
  roles:
    - role: openstack-interfaces
      tags:
        - openstack-interfaces

Ansible Role

I’ve already spoiled the name of the role we’ll create in the playbook above.
I’ll create the directory structure for it as usual.

$ mkdir -p /etc/ansible/roles/openstack-interfaces/tasks/

Let’s start adding tasks.

Packages

Although the OpenStack-Ansible documentation includes a list of required packages, we’ll not be installing all of them in this role.
In fact, the only package we are going to include from the documentation is “bridge-utils”, which allows the configuration of network bridges.

We’ve gone over installing packages in previous posts, so I won’t go into too much detail now.
Here’s the first task in our file “/etc/ansible/roles/openstack-interfaces/tasks/main.yml”

- name: Install prerequisite packages
  apt:
    name: bridge-utils
    state: present
    cache_valid_time: 3600

Let’s move on to our configuration.

Interfaces

Unlike our Cumulus switches, where our interfaces are configured via the file “/etc/network/interfaces” and brought up with the “networking” service, Ubuntu 18.04 implements a different tool for configuration called Netplan and brings this configuration to fruition through the newer “systemd-networkd” service.

Netplan is just another layer of configuration abstraction that sits on top of existing linux networking tools.
What this means for us is that we’re moving where we define our network interface configuration (“/etc/netplan/”), changing the format of this configuration to the more readable YAML (yes, the same configuration language that Ansible uses) and then calling Netplan to have this applied, through the appropriate tool, onto the network.

Let’s take a look at some example configuration that is already being used on our hosts.
The below configuration exists in the file “/etc/netplan/01-netcfg.yaml”, which is the default configuration file for Ubuntu 18.04 hosts:

# This file describes the network interfaces available on your system
# For more information, see netplan(5).
network:
  version: 2
  renderer: networkd
  ethernets:
    enp0s3:
      dhcp4: yes

Pretty simple for the entire starting network configuration of our host right?
Let’s go over some core parts before we start our own configuration:

  • network: Top-level node. Always present.
  • version: The version of YAML being used.
  • renderer: The network daemon that Netplan will configure. At the time of writing two options are available - NetworkManager (which is what our Cumulus switches use) and Systemd-networkd (which is what our Ubuntu hosts use by default).
  • ethernets: A block specifying the type of interfaces being configured, under which new interfaces can be defined. Other types include wifis, bridges, bonds, tunnels and vlans.

 

Now that I’ve covered how Netplan works a little, I think it’s best if we dive right into our Ansible task and handlers.

Netplan configuration files live in “/etc/netplan” (as seen above) and must end in “.yaml” to be picked up.

---
- name: Configure Interfaces
  template:
    src: "interfaces_custom.j2"
    dest: "/etc/netplan/interfaces_custom.yaml"
  notify:
    - Apply Netplan configuration

This requires the presence of a handler file (“handlers/main.yml”) to run the command “netplan apply”, which tells Netplan to put our configuration file into action.

---
- name: Apply Netplan configuration
  command: netplan apply

I’ll cover the content of the template file in two parts, starting with the configuration of the loopback and physical interfaces on our hosts.

Loopback and Physical Interfaces

As with our Cumulus switches, our loopback and physical ports are there to support the routing topology over which OpenStack will communicate.
Our requirements for configuration will also remain the same, with our loopback interface needing an IP address so that BGP peering relationships can be established; and our physical interfaces using the same address for OSPF unnumbered (which I covered in my previous post).

I’m going to mix it up this time and define our variables before diving into our template.
We’re going to want to define two items as variables for both hosts - our loopback IP addresses and physical interface numbers.

Because our hosts share the same physical port layout (through our Vagrant/VirtualBox configuration), they both end up with the same network interface names for their physical ports (enp0s8 and enp0s9). We can, therefore, use the same variable for both of our hosts.
I’ll create this variable in the new file “/etc/ansible/group_vars/openstack_hosts/vars.yml”:

---
openstack_hosts_network_interfaces:
  - enp0s8
  - enp0s9

# OSPF Routing
openstack_hosts_routing_ospf_unnumbered: True
openstack_hosts_routing_ospf_enabled: True

# BGP Routing
openstack_hosts_routing_bgp_enabled: True
openstack_hosts_routing_bgp_evpn_enabled: True

You’ll see that I’ve also created some other variables that we used in our Cumulus interface and routing configurations. These let us say whether we’re using a given protocol on our hosts.

We also need to define the IP address that each host will use on its’ loopback interface.
Here’s what “/etc/ansible/host_vars/openstack-compute/vars.yml” looks like as an example:

---
openstack_host_loopback_address: "192.168.11.132/32"

 

Now that we’ve set ourselves up with some variables, let’s begin writing our template (“templates/interfaces_custom.j2”):

network:
  version: 2
  renderer: networkd
  ethernets:
    #######################
    # Loopback Interfaces #
    #######################
    lo:
      match:
        name: lo
{% if openstack_host_loopback_address is defined %}
      addresses:
        - {{ openstack_host_loopback_address }}
{% endif %}
    #######################
    # Physical Interfaces #
    #######################
{% for port in openstack_hosts_network_interfaces | sort %}
    {{ port }}:
{% if openstack_hosts_routing_ospf_unnumbered == true %}
      addresses:
        - {{ openstack_host_loopback_address }}
{% endif %}
      mtu: {{ openstack_lab_mtu }}
{% endfor %}

This template starts with the same configuration lines as the default one that we looked at above.
After we specify the interface type with “ethernets:”, we can define the interfaces that we want to bring up underneath.

The first of which is our loopback interface.
Because we are simply adding an IP address to an existing interface (“lo”), Netplan needs us to tell which interface we are adding configuration to.
This is done with the “match” and “name” keys, essentially telling Netplan to configure the interface that already exists with the name of “lo”.
We then use the “addresses” key followed by a list (of one in our case) of IP addresses that we want configured.

Our physical interfaces are very similar. Define the interface name, give it an IP address and then configure the MTU by using the “mtu” key.

OpenStack Bridges

As I eluded to before, I’m also going to include in this template the configuration of the network bridges required by OpenStack-Ansible.
This is simply so that we don’t need to configure this file again in another role.

OpenStack-Ansible utilises containers to operate the many services that are included in an OpenStack installation. To connect to the outside world, the containers need to talk via bridges that need to be set up on the host before deployment.
These bridges are as follows:

  • br-mgmt: Provides management of and communication between the infrastructure and OpenStack services.
  • br-storage: Provides access between OpenStack services and storage devices.
  • br-vlan: Provides internal communication between tenant networks that are VLAN tagged or flat (no VLAN tag) and external communication to outside of the OpenStack network through “Provider networks” (NAT).
  • br-vxlan: Only required if the environment is configured create virtual networks using VXLAN. Provides internal communication between tenant networks that use VXLAN tunnelling.

The documentation helps out here, by showing which bridges need to be present on a given host.
For our lab deployment, both hosts are going to be acting as network (control), compute and storage nodes.
This translates to needing all 4 bridges (br-mgmt, br-storage, br-vxlan and br-vlan) on both hosts, with IP addresses on all except br-vlan.

 

Like in our above step, let’s define variables first.
Because the IP addresses that we configure on these bridges are going to be different on each host, let’s define the variables under the “host_vars” hierarchy.
Here’s what I’ve added to “/etc/ansible/host_vars/openstack-compute/vars.yml” as an example:

openstack_host_network_bridges:
  mgmt: "10.131.16.1/24"
  storage: "10.131.17.1/24"
  vxlan: "10.131.18.1/24"
  vlan:

Note the blank value for the “vlan” key, because that bridge will not have an IP address configured.

 

These variables come together to form the below addition to our interface configuration template:

###########################
# Local Bridge Interfaces #
###########################
bridges:
{% for bridge, address in openstack_host_network_bridges.items() | sort %}
  {{ bridge }}:
    interfaces: []
    parameters:
      stp: false
      forward-delay: 0
{% if address is not none %}
    addresses:
      - {{ address }}
{% endif %}
{% endfor %}

Most of this configuration was done by looking at OpenStack-Ansible’s examples and translating this into to the Netplan syntax for configuration of bridges.

Let’s walk through the configuration of each bridge:

  • bridges: - Tells Netplan that the following interfaces are bridges.
  • {{ bridge }} - For every key that exists in the variable “openstack_host_network_bridges” for the host, create an interface with the name of the key with the following configuration underneath.
  • interfaces: [] - Usually we would assign network interfaces that are participating in the bridge here, but instead we say that there are none. OpenStack-Ansible is going to add ports to this bridge when it creates the appropriate containers.
  • parameters: - Tells Netplan to set special bridge options, depending on the nested keys and values.
  • stp: false - The first option we use to turn spanning tree off . This is because the bridges aren’t uplinking to any other switches (rather, this bridge will be advertised as a prefix by iBGP for reachability between hosts).
  • forward-delay: 0 - The second option says that there is to be no forwarding delay on the bridge (Tells the bridge to forward traffic immediately instead of going through the usual listening and learning stages).
  • If the bridge has an IP address specified, we set this using the standard syntax.

As simple as that, we’ve configured all prerequisite network interfaces.

Vagrant

The last thing we need to do is tell Vagrant that it needs to run the playbook after it has set up both OpenStack hosts.
I’ll accomplish this by including the provision block after the last OpenStack host, just like we did with the Cumulus provisioning block.

config.vm.define "openstack_compute" do |device|
  ...
  # Ansible Playbook for all OpenStack hosts
  device.vm.provision :ansible do |ansible|
    ansible.inventory_path = "/etc/ansible/hosts"
    ansible.playbook = "/etc/ansible/playbooks/openstack-hosts.yml"
    ansible.limit = "openstack_hosts"
  end
end

Testing

After all of our configuration, when we run vagrant up on our local machine, we get the following output after it has created the two OpenStack hosts:

$ vagrant up
...
==> openstack_compute: Running provisioner: ansible...
    openstack_compute: Running ansible-playbook...

PLAY [Configuring OpenStack hosts] *********************************************

TASK [Gathering Facts] *********************************************************
ok: [openstack-control]
ok: [openstack-compute]

TASK [openstack-interfaces : Install prerequisite packages] ********************
changed: [openstack-compute]
changed: [openstack-control]

TASK [openstack-interfaces : Configure Interfaces] *****************************
changed: [openstack-control]
changed: [openstack-compute]

RUNNING HANDLER [openstack-interfaces : Apply Netplan configuration] ***********
changed: [openstack-compute]
changed: [openstack-control]

PLAY RECAP *********************************************************************
openstack-compute          : ok=1    changed=3    unreachable=0    failed=0   
openstack-control          : ok=1    changed=3    unreachable=0    failed=0

Looks like everything was configured, let’s check that it’s taken effect on the hosts.

Interfaces

Using vagrant ssh $host we can connect to both of our hosts and find out whether everything looks as it should.

Firstly, let’s check whether our bridges exist:

$ brctl show
bridge name   bridge id           STP enabled   interfaces
mgmt          8000.ce995af87981   no		
storage       8000.864182527489   no		
vlan          8000.daeb42cdd17b   no		
vxlan         8000.000000000000   no

Although the command doesn’t show a lot of information (yet - brctl becomes a very useful command once MAC addresses are being learnt by each bridge), it does confirm that we’ve created the bridges on the host.

We can confirm this with the command ip address show:

$ ip address show | grep "UP\|inet" | grep -v "inet6"
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1
    inet 127.0.0.1/8 scope host lo
    inet 192.168.11.132/32 brd 192.168.11.132 scope global lo:1
2: enp0s3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP group default qlen 1000
    inet 192.168.11.232/24 brd 192.168.11.255 scope global enp0s3
3: enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9000 qdisc pfifo_fast state UP group default qlen 1000
    inet 192.168.11.132/32 brd 192.168.11.132 scope global enp0s8
4: enp0s9: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9000 qdisc pfifo_fast state UP group default qlen 1000
    inet 192.168.11.132/32 brd 192.168.11.132 scope global enp0s9
5: vxlan: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN group default qlen 1000
    inet 172.16.2.132/24 brd 172.16.2.255 scope global vxlan
6: vlan: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN group default qlen 1000
7: storage: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN group default qlen 1000
    inet 172.16.1.132/24 brd 172.16.1.255 scope global storage
8: mgmt: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN group default qlen 1000
    inet 172.16.0.132/24 brd 172.16.0.255 scope global mgmt

This command shows us each interface present and its’ associated IP address.

Conclusion

In this post, we covered some basic network interface configuration on our OpenStack hosts, including the bridges that OpenStack-Ansible needs present.

In the next post, we’ll build on top of this configuration by bringing up OSPF and iBGP.

Appendices

References

Ansible - “template” module
Ansible - “service” module
Ansible - “command” module
Ansible - “apt” module
Netplan - Reference Documentation
OpenStack Ansible - Deployment Guide
OpenStack Ansible - Routed Environment Example

Versions used

Desktop Machine: kubuntu-18.04
VirtualBox: virtualbox-5.2.18
Vagrant: 2.2.3
Cumulus VX Vagrant Box: CumulusCommunity/cumulus-vx (virtualbox, 3.7.2)
Ubuntu Server Vagrant Box: geerlingguy/ubuntu1804 (virtualbox, 1.0.6)
Ansible: 2.7.6