This is the documentation for the latest development version of Antrea. Both code and docs may be unstable, and these docs are not guaranteed to be up to date or correct. See the latest version.
Antrea supports chaining with routed CNI implementations such as EKS CNI. In this mode, Antrea enforces Kubernetes NetworkPolicy, and delegates Pod IP management and network connectivity to the primary CNI.
Antrea is designed to work as NetworkPolicy plug-in to work together with a routed CNIs. For as long as a CNI implementation fits into this model, Antrea may be inserted to enforce NetworkPolicy in that CNI's environment using Open VSwitch(OVS).
In addition, Antrea working as NetworkPolicy plug-in automatically enables Antrea-proxy, because it requires Antrea-proxy to load balance Pod-to-Service traffic.
The above diagram depicts a routed CNI network topology on the left, and what it looks like after Antrea inserts the OVS bridge into the data path.
The diagram on the left illustrates a routed CNI network topology such as AWS EKS. In this topology a Pod connects to the host network via a point-to-point(PtP) like device, such as (but not limited to) a veth-pair. On the host network, a host route with corresponding Pod's IP address as destination is created on each PtP device. Within each Pod, routes are configured to ensure all outgoing traffic is sent over this PtP device, and incoming traffic is received on this PtP device. This is a spoke-and-hub model, where to/from Pod traffic, even within the same worker Node must traverse first to the host network and be routed by it.
When the container runtime instantiates a Pod, it first calls the primary CNI to configure Pod's
IP, route table, DNS etc, and then connects Pod to host network with a PtP device such as a
veth-pair. When Antrea is chained with this primary CNI, container runtime then calls
Antrea Agent, and the Antrea Agent attaches Pod's PtP device to the OVS bridge, and moves the host
route to the Pod to local host gateway(
antrea-gw0) interface from the PtP device. This is
illustrated by the diagram on the right.
Antrea needs to satisfy that
antrea-gw0in the host network, are received by the Pods exactly the same as if the OVS bridge had not been inserted.
To satisfy the above requirements, Antrea needs no knowledge of Pod's network configurations nor of underlying CNI network, it simply needs to program the following OVS flows on the OVS bridge:
These flows together handle all Pod traffic patterns.