CNI with Calico

Calico is a software-defined network solution that can be used with Kubernetes. Support for Calico in Charmed Kubernetes is provided in the form of a calico subordinate charm.

Unlike Flannel, Calico provides out-of-the-box support for the NetworkPolicy feature of Kubernetes, along with different modes of network encapsulation that advanced users may find useful for optimising the throughput of their clusters.

Cloud configuration

Calico’s network traffic is filtered on many clouds, and may require special configuration at the cloud level to support it.

If Calico is configured to use BGP mode (the default), then all of the Kubernetes instances must be located in the same subnet for Calico to work out of the box. Alternatively, the Calico charm may be configured with external BGP peers to allow cross-subnet traffic to work.

If Calico is configured to use IPIP mode, then the cloud must be configured to allow IPIP (protocol 4) network traffic.

AWS

On AWS, it is recommended to run Calico in BGP mode. This usually requires the creation of a VPC that has only a single subnet associated with it. See Juju’s Creating an AWS VPC documentation for instructions on how to create a VPC that’s compatible with Juju.

After deployment, each AWS instance must be manually configured to disable source/destination checks. See AWS’s Disabling Source/Destination Checks documentation for instructions.

Deploying Charmed Kubernetes with Calico

To deploy CDK with Calico, deploy the kubernetes-calico bundle:

juju deploy cs:~containers/kubernetes-calico

The calico bundle is identical to the standard charmed-kubernetes bundle with the exception of replacing flannel with calico. You can apply any customisation overlays that would apply to charmed-kubernetes to this bundle also.

Calico configuration options

Checking the current configuration

To check the current configuration settings for Calico, run the command:

juju config calico

Setting a config option

To set an option, simply run the config command with and additional <key>=<value> argument. For example, to disable NAT on outgoing traffic:

juju config calico nat-outgoing=False

Config settings which require additional explanation are described below.

Calico IPIP configuration

By default, IPIP encapsulation is disabled. To enable IPIP encapsulation, set the ipip charm config to Always:

juju config calico ipip=Always

Alternatively, if you would like IPIP encapsulation to be used for cross-subnet traffic only, set the ipip charm config to CrossSubnet:

juju config calico ipip=CrossSubnet

Calico BGP configuration

The default configuration of Calico uses BGP mode, with all Calico nodes connected in a full node-to-node mesh, and with no external peerings. This comes with some limitations which will be explained in the following sections.

BGP with multiple subnets

If BGP mode is in use, and Calico units are deployed to separate subnets, then Calico must be configured to peer with external routers in order to accommodate cross-subnet traffic.

For example, if you have units across two subnets (10.0.0.0/24 and 10.0.1.0/24), and a single router that connects them (with IPs 10.0.0.1, 10.0.1.1, and AS number 64512), then the simplest configuration would be to configure Calico to peer with that router and to use the same AS number.

Configure the AS number:

juju config calico global-as-number=64512

And configure the external peering:

juju config calico subnet-bgp-peers="
10.0.0.0/24:
- address: 10.0.0.1
  as-number: 64512
10.0.1.0/24:
- address: 10.0.1.1
  as-number: 64512
"

You will also need to configure the router to peer with each Calico node. This step varies based on the router’s BGP implementation. Here’s an example configuration for BIRD.

BGP route reflection

By default, all Calico nodes are connected in a full BGP mesh. This works well for small deployments, but does not scale well to larger deployments. For large deployments, it is recommended that you configure Calico to use route reflection.

For example, in a deployment with more than 10 Calico units, you might configure it such that units calico/0 (with IP 10.0.0.2) and calico/1 (with IP 10.0.0.3) are route reflectors, and all other Calico units only peer with them. To do this, you would first configure route reflector cluster IDs:

juju config calico route-reflector-cluster-ids="
0: 0.0.0.0
1: 0.0.0.0
"

Then configure every node to peer with the route reflectors:

juju config calico global-bgp-peers="
- address: 10.0.0.2
  as-number: 64512
- address: 10.0.0.3
  as-number: 64512
"

And disable the node-to-node-mesh:

juju config calico node-to-node-mesh=false

BGP with the AS Per Rack model

One commonly used model for Calico deployments on bare metal is the AS Per Rack model. The configuration in this scenario can be tricky, so let’s walk through an example.

Given:

• Two racks:
  • Rack 0
    • Subnets 10.0.0.0/24, 10.0.1.0/24
    • ToR router with IPs 10.0.0.1, 10.0.1.1
    • Desired route reflectors:
      • calico/0 at 10.0.0.2
      • calico/1 at 10.0.1.2
  • Rack 1
    • Subnets 10.0.2.0/24, 10.0.3.0/24
    • ToR router with IPs 10.0.2.1, 10.0.3.1
    • Desired route reflectors:
      • calico/2 at 10.0.2.2
      • calico/3 at 10.0.3.2

Assign a different AS number to each rack:

juju config calico subnet-as-numbers="
10.0.0.0/24: 64512
10.0.1.0/24: 64512
10.0.2.0/24: 64513
10.0.3.0/24: 64513
"

Next, configure the route reflectors. Each route reflector needs to be assigned a cluster ID. The exact cluster ID used isn’t important, as long as both route reflectors on a given rack share the same cluster ID. For this example, we’ll choose 0.0.0.0 for rack 0 and 0.0.0.1 for rack 1.

Configure the route reflectors:

juju config calico route-reflector-cluster-ids="
0: 0.0.0.0
1: 0.0.0.0
2: 0.0.0.1
3: 0.0.0.1
"

Configure route reflectors to peer with the top of rack switches:

juju config calico unit-bgp-peers="
0:
- address: 10.0.0.1
  as-number: 64512
1:
- address: 10.0.1.1
  as-number: 64512
2:
- address: 10.0.2.1
  as-number: 64513
3:
- address: 10.0.3.1
  as-number: 64513
"

Configure Calico nodes to peer with the route reflectors:

juju config calico subnet-bgp-peers="
10.0.0.0/24:
- address: 10.0.0.2
  as-number: 64512
- address: 10.0.1.2
  as-number: 64512
10.0.1.0/24:
- address: 10.0.0.2
  as-number: 64512
- address: 10.0.1.2
  as-number: 64512
10.0.2.0/24:
- address: 10.0.2.2
  as-number: 64513
- address: 10.0.3.2
  as-number: 64513
10.0.3.0/24:
- address: 10.0.2.2
  as-number: 64513
- address: 10.0.3.2
  as-number: 64513
"

And disable the node-to-node mesh:

juju config calico node-to-node-mesh=false

You will also need to configure the top of rack switches to peer with the Calico route reflectors. Instructions for doing this should be provided in the documentation for the specific model of switch you are using.

Configuring multiple Calico pools

The Calico charm creates a single IPPool. If multiple IPPools pools are desired, then you must first disable the charm’s built-in pool management:

juju config calico manage-pools=false

Then use calicoctl to create your own pools. For example:

juju ssh calico/0 calicoctl apply -f - << EOF
apiVersion: projectcalico.org/v3
kind: IPPool
metadata:
  name: my-pool
spec:
  cidr: 192.168.0.0/24
  ipipMode: Never
  natOutgoing: true
EOF

Using a private Docker registry

For a general introduction to using a private Docker registry with Charmed Kubernetes, please refer to the Private Docker Registry page.

In addition to the steps documented there, you will need to upload the following images to the registry:

docker.io/calico/node:v3.6.1
docker.io/calico/kube-controllers:v3.6.1

And configure Calico to use the registry:

export IP=`juju run --unit docker-registry/0 'network-get website --ingress-address'`
export PORT=`juju config docker-registry registry-port`
export REGISTRY=$IP:$PORT
juju config calico \
  calico-node-image=$registry/calico/node:v3.6.1 \
  calico-policy-image=$registry/calico/kube-controllers:v3.6.1

Troubleshooting

If there is an issue with connectivity, it can be useful to inspect the Juju logs. To see a complete set of logs for Calico

juju debug-log --replay --include=calico

For additional troubleshooting pointers, please see the dedicated troubleshooting page.

Useful links

• The Calico website has a thorough explanation of its network management strategy.