Integration with Fuel
Calico plugins are available for Fuel 6.1 and 7.0, and work is in progress for Fuel 9. Fuel plugin code for Calico is at http://git.openstack.org/cgit/openstack/fuel-plugin-calico.
The plugin for Fuel 7.0 is currently undergoing final review and certification; you can find the plugin code at git.openstack.org, and its documentation in pending changes on review.openstack.org:
- Code: https://git.openstack.org/cgit/openstack/fuel-plugin-calico/log/?h=7.0
- User Guide: https://review.openstack.org/#/c/281239/
- Test Plan and Report: https://review.openstack.org/#/c/282362/
The rest of this document describes our integration of Calico with Mirantis Fuel 6.1. It is presented in sections covering the following aspects of our integration work.
- Objective: The system that we are aiming to deploy.
- Cluster Deployment: The procedure to follow to deploy such a system.
- Calico Demonstration: What we recommend for demonstrating Calico networking function.
- Detailed Observations: Some further detailed observations about the elements of the deployed system.
We will deploy an OpenStack cluster with a controller node and n compute nodes, with Calico providing the network connectivity between VMs that are launched on the compute nodes.
The key components on the controller node will be:
- the standard OpenStack components for a controller, including the Neutron service
- the Calico/OpenStack Plugin, as a Neutron/ML2 mechanism driver
- a BIRD instance, acting as BGP route reflector for the compute nodes.
The key components on each compute node will be:
- the standard OpenStack components for a compute node
- the Calico Felix agent
- a BIRD instance, running BGP for that compute node.
IP connectivity between VMs that are launched within the cluster will be established by the Calico components, according to the security groups and rules that are configured for those VMs through OpenStack.
The procedure for deploying such a cluster consists of the following steps.
- Prepare a Fuel master (aka admin) node in the usual way.
- Install the Calico plugin for Fuel on the master node.
- Deploy an OpenStack cluster in the usual way, using the Fuel web UI.
The following subsections flesh out these steps.
Prepare a Fuel master node
Install the Calico plugin for Fuel on the master node
The Calico plugin has been certified by Mirantis and is available for download from the Fuel Plugin Catalog. Alternatively, you can build a copy of the plugin yourself, following the instructions on the plugin’s GitHub page.
However you obtain a copy of the Calico plugin, you will need to copy it onto the master node and install it with:
fuel plugins --install calico-fuel-plugin-<version>.noarch.rpm
You can check that the plugin was successfully installed using:
fuel plugins --list
Deploy an OpenStack cluster
Use the Fuel web UI to deploy an OpenStack cluster in the usual way, with the following guidelines.
- Create a new OpenStack environment, selecting:
- Juno on Ubuntu Trusty (14.x)
- “Neutron with VLAN segmentation” as the networking setup
- Under the settings tab, make sure the following options are checked:
- “Assign public network to all nodes”
- “Use Calico Virtual Networking”
- Network settings as advised in the following subsections.
- Add nodes (for meaningful testing, you will need at least two compute nodes in addition to the controller).
- Deploy changes.
Fuel assigns a ‘public’ IP address, from the range that you specify here, to each node that it deploys. It also creates an OpenStack network with this subnet, and uses that for allocating floating IPs.
Therefore these IP addresses exist to allow access from within the cluster to the outside world, and vice versa, and should probably be routable from the wider network where the cluster is deployed.
For the purposes of this document we’ll use the 172.18.203.0/24 range of public addresses: feel free to change this to match your own local network.
- IP Range: 172.18.203.40 - 172.18.203.49
- CIDR: 172.18.203.0/24
- Use VLAN tagging: No
- Gateway: 172.18.203.1
- Floating IP ranges: 172.18.203.50 - 172.18.203.59
By default, Fuel associates the public IP address with the second NIC
eth1) on each node.
Fuel assigns a ‘management’ IP address, from the range that you specify here, to each node that it deploys. These are the addresses that the nodes within the cluster use to communicate with each other. For example, nova-compute on each compute node communicates with the Neutron server on the controller node by using the controller node’s management address.
- CIDR: 192.168.0.0/24
- Use VLAN tagging: Yes, 101
By default, Fuel associates the management IP address with the first NIC
eth0) on each node.
With Calico networking, in addition:
- BGP sessions are established, between BIRD instances on the compute nodes and on the route reflector, using these management IP addresses
- Data between VMs on different compute nodes is routed using these
management IP addresses, which means that it flows via the compute
Storage networking is not needed for a simple OpenStack cluster. We left the following settings as shown, and addresses from the specified range are assigned, but are not used in practice.
- CIDR: 192.168.1.0/24
- Use VLAN tagging: Yes, 102
Neutron L2 Configuration
Neutron L2 Configuration is not needed in a Calico system, but we have left the following settings as shown, as we have not yet had time to simplify the web UI for Calico networking.
- VLAN ID range: 1000 - 1030
- Base MAC address: fa:16:3e:00:00:00
Neutron L3 Configuration
Neutron L3 Configuration is not needed in a Calico system, but we have left the following settings as shown, as we have not yet had time to simplify the web UI for Calico networking.
- Internal network CIDR: 192.168.111.0/24
- Internal network gateway: 192.168.111.1
- DNS servers: 220.127.116.11, 18.104.22.168
Check BGP connectivity on the controller
Once the deployment is complete, you may wish to verify that the route reflector running on the controller node has established BGP sessions to all of the compute nodes.
To do this, log onto the controller node and run:
birdc show protocols all
To demonstrate Calico networking, please run through the following steps.
In the OpenStack web UI, under Project, Network, Networks, create a network and subnet from which instance IP addresses will be allocated. We use the following values.
- Name: ‘demo’
- IP subnet: 10.65.0.0/24
- Gateway: 10.65.0.1
- DHCP-enabled: Yes.
Under Project, Compute, Access & Security, create two new security groups. For each security group, select ‘Manage Rules’ and add two new rules:
- Allow incoming ICMP (ping) traffic only if it originates from other
instances in this security group:
- Rule: ALL ICMP
- Direction: Ingress
- Remote: Security Group
- Security Group: Current Group
- Ether Type: IPv4
- Enable SSH onto instances in this security group:
- Rule: SSH
- Remote: CIDR
- CIDR: 0.0.0.0/0
Under Project, Instances, launch a batch of VMs – enough of them to ensure that there will be at least one VM on each compute node – with the following details.
- Flavor: m1.tiny
- Boot from image: TestVM
- Under the Access & Security tab, select one of your new security groups (split your instances roughly 50:50 between the two security groups).
- Under the Networking tab, drag ‘demo’ into the ‘Selected Networks’ box.
Under Admin, Instances, verify that:
- the requested number of VMs (aka instances) has been launched
- they are distributed roughly evenly across the available compute hosts
- they have each been assigned an IP address from the range that you configured above (e.g. 10.65.0/24)
- they reach Active status within about a minute.
Log on to one of the VMs, e.g. by clicking on one of the instances and then on its Console tab, and use ‘ping’ to verify connectivity is as expected from the security group configuration, i.e. that you can ping the IP addresses of all of the other VMs in the same security group, but you cannot ping the VMs in the other security group.
Note that whilst the VMs should be able to reach other (security group configuration permitting), they are not expected to have external connectivity unless appropriate routing has been set up:
- For outbound access, you need to ensure that your VMs can send traffic to your border gateway router (typically this will be the case, because usually your compute hosts will be able to do so). The border gateway can then perform SNAT.
- For inbound connections, you need assign a publically routable IP address to your VM - that is, attach it to a network with a public IP address. You will also need to make sure that your border router (and any intermediate routers between the border router and the compute host) can route to that address too. The simplest way to do that is to peer the border router with the route reflector on the controller.
This section records some more detailed notes about the state of the cluster that results from following the above procedure.
Reading this section should not be required in order to demonstrate or understand OpenStack and Calico function, but it may be useful as a reference if a newly deployed system does not appear to be behaving correctly.
Elements required for Calico function
This subsection records elements that are required for Calico function, and that we have observed to be configured and operating correctly in the cluster.
On the controller:
- The BIRD BGP route reflector has established sessions to all the compute nodes.
- The Neutron service is running and has initialized the Calico ML2 mechanism driver.
On each compute node:
- The Calico Felix agent is correctly configured, and running.
- There is an established BGP session to the route reflector on the controller.
Elements not required for Calico function, but benign
This subsection records elements that are not required for Calico function, but that we have observed to be operating in the cluster. These all result from the fact that the procedure first deploys a traditional Neutron/ML2/OVS cluster, and then modifies that to use Calico instead of OVS, but does not clean up all of the OVS-related elements.
We believe that all of these elements are benign, in that they don’t obstruct or fundamentally change the Calico networking behavior. However it would be better to remove them so as to clarify the overall picture, and maybe to improve networking performance. We plan to continue working on this.
On the controller:
- Various Neutron agents are running that Calico does not require.
On each compute node:
- Two Neutron agents are running that Calico does not require.
- There is a complex set of OVS bridges present, that Calico does not require.