Configuring MTU

Depending on the environment Calico is being deployed into it may be helpful or even necessary to configure the MTU of the veth (or TAP) that is attached to each workload and the tunnel devices if IP-in-IP or VXLAN is enabled.

Selecting MTU size

Since MTU is a global property of the network path between endpoints, the MTU for workloads needs to be set to the minimum MTU of any path that packets may take.

If you are using an overlay such as IP-in-IP or VXLAN, the extra overlay header used by those protocols reduces the minimum MTU by the size of the header. IP-in-IP uses a 20-byte header, VXLAN uses a 50-byte header. Hence,

• If you use VXLAN anywhere in your pod network, you should select an MTU which is network MTU minus 50.
• If you do not use VXLAN but you do use IP-in-IP, you should select an MTU which is network MTU minus 20.

You should set the workload endpoint MTU and the tunnel MTUs to the same value. This is so that all paths have the same MTU.

Common MTU sizes

Default MTU sizes

The default MTU for workload interfaces is 1500, this is to match the most common network MTU size. The default MTU for the IP-in-IP tunnel device is 1440 to match the value needed in GCE. Similarly, the default for VXLAN is 1410.

Using flannel for networking

When using flannel for networking, the MTU for the network interfaces should match the MTU of the flannel interface. In the above table the 4th column “Calico MTU with VXLAN” is the expected MTU when using flannel configured with VXLAN.

MTU configuration

It is the job of the network plugin to create new workload interfaces. The CNI plugin, which is used by Kubernetes, supports configuring the MTU of the workload interface through the CNI configuration file.

The user will also want to configure Calico’s IP-in-IP/VXLAN interface MTU when IP-in-IP/VXLAN is enabled on the cluster. Refer to the MTU table at the top of the page to choose the value that matches your environment.

Note: The MTU on existing workloads will not be updated with these changes. To update workload MTUs, see the section that corresponds to your plugin type.

MTU configuration with CNI

MTU is set in the by the "mtu": <MTU size> field of the CNI configuration. Example:

  {
    "name": "any_name",
    "cniVersion": "0.1.0",
    "type": "calico",
    "mtu": 1480,
    "ipam": {
      "type": "calico-ipam"
    }
  }

Note: If using Kubernetes self-hosted manifests, you should modify the veth_mtu value in the Calico ConfigMap instead, and leave "mtu" here set to __CNI_MTU__. See below for more details.

MTU configuration with Kubernetes self-hosted manifests

When using Kubernetes self-hosted manifests, the CNI plugin derives the MTU value from the veth_mtu field of the calico-config ConfigMap, and it is set to 1440 by default. On restart of the calico/node pods, any references to __CNI_MTU__ are replaced by the veth_mtu value and inserted into the CNI configuration file (aka conflist) at the directory specified by Kubernetes (currently defaults to /etc/cni/net.d/).

Restarting the calico/node pods will also update any Calico tunnel network interfaces on that node. From this point forward, any pods started will also have the updated MTU value.

Setting tunnel MTU with a Felix environment variable

Passing in the environment variable FELIX_IPINIPMTU (or FELIX_VXLANMTU) when running the calico/node container will set the MTU for Felix to use.

When using the Kubernetes self-hosted manifests, Felix derives the IP-in-IP value from the veth_mtu field of the calico-config ConfigMap, which is set to 1440 by default.

Setting tunnel MTU with calicoctl

To set the IP-in-IP MTU value for all Calico nodes in your cluster, use the following command to retrieve the current Felix settings.

calicoctl get felixconfig default --export -o yaml > felix.yaml

Modify ipipMTU (or vxlanMTU) to the intended integer value.

vim felix.yaml

Replace the current felixconfig settings.

calicoctl replace -f felix.yaml

Note: Setting the ipipMTU (of vxlanMTU) config option will result in an immediate update of the tunnel interface MTU on all of the active nodes in your cluster.