Balance innovation and agility with security and compliance
risks using a 3-step process across all cloud infrastructure.
Step up business agility without compromising
security or compliance
Everything you need to become a Kubernetes expert.
Always for free!
Everything you need to know about Magalix
culture and much more
Kubernetes is a technology that helps you get the most out of your hardware. Containers are deployed on several nodes, making sure that every CPU cycle, every byte of memory, and every block of storage is not wasted. However, this is no easy task. Several challenges must be addressed when designing how the cluster will handle networking among containers:
Kubernetes is a highly modular, open-source project. Several components were left to the community to develop. In particular, implementing a cluster-networking solution must conform to a set of high-level rules. They can be summarized as follows:
To appreciate the simplicity of this design, let’s see how we can manually create a number of containers (using Docker, for example) that will be distributed on a number of physical hosts, and how they can communicate without the Kubernetes design. First, you’ll need to use NAT to ensure that no port collision happens when more than one container tries to use the same port. Let’s say two Apache containers, both are running on port 80. None of those containers can receive traffic by exposing port 80 on the host; as a port collision will occur. This is only possible through NAT. Using NAT means that the container will not communicate through its own IP address or port. Rather, its IP will be hidden behind the NAT IP, and a unique port on the NAT interface (for example, 8080) will forward traffic to port 80 on the container. The second container will use the same NAT IP but with a different port and so on. The following graph depicts how Kubernetes implements its networking model versus the traditional way.
So, as you can see Kubernetes eliminates the need for NAT or link containers.
There are a number of networking models that adhere to the above rules. In this article, we’ll select some of them for discussion. But, before listing the different network plugin examples, let’s have a quick overview of some important Kubernetes networking terms.
In general, we can define networks as underlay and overlay types:
The Underlay network is closer to the physical layer. It includes switches, routers, VLANs, and so on. It is the basis on which overlay networks are built. It tends to be less scalable due to technical limitations. However, since it’s closer to the actual hardware, it is slightly faster than an overlay.
Overlay network refers to the virtual network layer. In this type, you’ll hear terms like veth (virtual eth or virtual network interface), and VxLAN. It is designed to be highly scalable than the underlying network. For example, while VLANs in the underlying network support only 4096 identifiers, VxLAN can reach up to 16 million ones.
Kubernetes supports both networking models, so you can base your model of choice on other factors than whether or not the cluster can handle it.
A CNI is simply a link between the container runtime (like Docker or rkt) and the network plugin. The network plugin is nothing but the executable that handles the actual connection of the container to or from the network, according to a set of rules defined by the CNI. So, to put it simply, a CNI is a set of rules and Go libraries that aid in container/network-plugin integration.
All of the CNIs can be deployed by simply running a pod or a daemonset that launches and manages their daemons. Let’s have a look now at the most well-known Kubernetes networking solutions.
Calico is a scalable and secure networking plugin. It can be used to manage and secure network policies not only for Kubernetes, but also for containers, virtual machines, and even bare metal servers. Calico works on Layer 3 of the network stack. It works by implementing a vRouter (as opposed to a vSwitch) on each node. Since it is working on L3, it can easily use the Linux kernel’s native forwarding functionality. The Felix agent is responsible for programming L3 Forwarding Information base with the IP addresses of the pods scheduled on the node where it is running.
Calico uses vRouters to allow pods to connect to each other across different nodes using the physical network (underlay). It does not use overlay, tunneling or VRF tables. Also, it does not require NAT since each pod can be assigned a public IP address that is accessible from anywhere as long as the security policy permits it.
Deployment differs based on the type of environment or the cloud provider where you’ll be hosting your cluster. This document contains all the supported Calico deployment methods.
Cilium uses layers 3, 4 (network), and layer 7 (application) to function. It brings a solution that is not only aware of the packets that pass through, but also the application and protocol (for example, HTTP) that those packets are using. Having such a level of inspection allows Cilium to control and enforce network and application security policies. Be aware, though that for this plugin to work, you must be using a Linux kernel that is equal to or higher than 4.8. That’s because Cilium uses a new kernel feature Berkeley Packet Filter (BPF), which can replace iptables.
Cilium runs a daemon called cilium-agent on each node. It compiles the BPF filters and transfers them to the kernel for further processing.
For different ways to install Cilium, which includes using your own machine (through minikube or microk8s), please follow this guide.
Weave Net is an easy-to-use, resilient, and fast-growing network plugin that can be used for more than just container networking. When installed, Weave Net creates a virtual router on each host (called peer). Those routers start communicating with each other to establish a protocol handshake and, later, learn the network topology. The plugin also creates a bridge interface on each host. All pods get attached to this interface, and they are assigned IP addresses and netmasks. Within the same node, Weave Net uses the kernel to move packets from one pod to another. This protocol is called the fast data path. When the packet is destined to a pod on another host, the plugin uses the sleeve protocol, in which UDP is used to contact the router on the destination host to transfer packets. Subsequently, those packets are captured by the kernel and passed to the target pod.
One way to install Weave Net on a Kubernetes cluster is to apply a daemonset which will automatically install the necessary containers for running the plugin on each node. Once up and running, all pods will use this network for their communication. The peers are self-configuring, so you can add more nodes to the cluster and they’ll use the same network without further configuration from your side.
Flannel is a networking plugin created by CoreOS. It implements cluster networking in Kubernetes by creating an overlay network. It starts a daemon called flanneld on each node. This daemon runs under a pod whose name starts with kube-flannel-ds-*. When assigning IP addresses, Flannel allocates a small subset of IPs of each host (by default, 10.244.X.0/24). This subset is brought from a larger, preconfigured address space. This subset is used to assign an IP address of each pod on the node.
Flannel uses the Kubernetes API server or the cluster’s etcd database directly to store information about the assigned subnets, network configuration, and the host IP address.
Packet forwarding among hosts is done through several protocols like UDP and VXLAN.
For the specific instructions for installing and running Flannel, please use this document.
In this article, we provided a gentle introduction to Kubernetes networking. The way Kubernetes was designed gives the user much freedom in choosing which components work together best (while still abiding by some governing rules). In the first part of this article, we started by explaining how Kubernetes designers thought about how networking should be implemented within the cluster.
Then we briefly discussed some of the Kubernetes networking concepts that help you understand the variety of networking plugins offered by the community. In the second part, we listed a sample of the most well-known networking plugins that are available for Kubernetes, how each of them was designed, and how they implement networking within the cluster.
Deciding on which network plugin to use with your coming project largely depends on what this project is, its network requirements, and the level and type of network security it needs to implement.
Self-service developer platform is all about creating a frictionless development process, boosting developer velocity, and increasing developer autonomy. Learn more about self-service platforms and why it’s important.
More and more businesses are adopting GitOps. Learn about the 5 reasons why GitOps is important for businesses.