Kubernetes approaches networking somewhat differently than Docker does by default. There are 4 distinct networking problems to solve:
Kubernetes assumes that pods can communicate with other pods, regardless of which host they land on. We give every pod its own IP address so you do not need to explicitly create links between pods and you almost never need to deal with mapping container ports to host ports. This creates a clean, backwards-compatible model where pods can be treated much like VMs or physical hosts from the perspectives of port allocation, naming, service discovery, load balancing, application configuration, and migration.
To achieve this we must impose some requirements on how you set up your cluster networking.
Before discussing the Kubernetes approach to networking, it is worthwhile to
review the “normal” way that networking works with Docker. By default, Docker
uses host-private networking. It creates a virtual bridge, called
default, and allocates a subnet from one of the private address blocks defined
in RFC1918 for that bridge. For each
container that Docker creates, it allocates a virtual ethernet device (called
veth) which is attached to the bridge. The veth is mapped to appear as
in the container, using Linux namespaces. The in-container
eth0 interface is
given an IP address from the bridge’s address range.
The result is that Docker containers can talk to other containers only if they are on the same machine (and thus the same virtual bridge). Containers on different machines can not reach each other - in fact they may end up with the exact same network ranges and IP addresses.
In order for Docker containers to communicate across nodes, they must be allocated ports on the machine’s own IP address, which are then forwarded or proxied to the containers. This obviously means that containers must either coordinate which ports they use very carefully or else be allocated ports dynamically.
Coordinating ports across multiple developers is very difficult to do at scale and exposes users to cluster-level issues outside of their control. Dynamic port allocation brings a lot of complications to the system - every application has to take ports as flags, the API servers have to know how to insert dynamic port numbers into configuration blocks, services have to know how to find each other, etc. Rather than deal with this, Kubernetes takes a different approach.
Kubernetes imposes the following fundamental requirements on any networking implementation (barring any intentional network segmentation policies):
What this means in practice is that you can not just take two computers running Docker and expect Kubernetes to work. You must ensure that the fundamental requirements are met.
This model is not only less complex overall, but it is principally compatible with the desire for Kubernetes to enable low-friction porting of apps from VMs to containers. If your job previously ran in a VM, your VM had an IP and could talk to other VMs in your project. This is the same basic model.
Until now this document has talked about containers. In reality, Kubernetes
applies IP addresses at the
Pod scope - containers within a
Pod share their
network namespaces - including their IP address. This means that containers
Pod can all reach each other’s ports on
localhost. This does imply
that containers within a
Pod must coordinate port usage, but this is no
different than processes in a VM. We call this the “IP-per-pod” model. This
is implemented in Docker as a “pod container” which holds the network namespace
open while “app containers” (the things the user specified) join that namespace
As with Docker, it is possible to request host ports, but this is reduced to a
very niche operation. In this case a port will be allocated on the host
and traffic will be forwarded to the
Pod itself is blind to the
existence or non-existence of host ports.
There are a number of ways that this network model can be implemented. This document is not an exhaustive study of the various methods, but hopefully serves as an introduction to various technologies and serves as a jumping-off point. If some techniques become vastly preferable to others, we might detail them more here.
For the Google Compute Engine cluster configuration scripts, we use advanced
assign each VM a subnet (default is
/24 - 254 IPs). Any traffic bound for that
subnet will be routed directly to the VM by the GCE network fabric. This is in
addition to the “main” IP address assigned to the VM, which is NAT’ed for
outbound internet access. A linux bridge (called
cbr0) is configured to exist
on that subnet, and is passed to docker’s
We start Docker with:
DOCKER_OPTS="--bridge=cbr0 --iptables=false --ip-masq=false"
This bridge is created by Kubelet (controlled by the
flag) according to the
Docker will now allocate IPs from the
cbr-cidr block. Containers can reach
each other and
Nodes over the
cbr0 bridge. Those IPs are all routable
within the GCE project network.
GCE itself does not know anything about these IPs, though, so it will not NAT
them for outbound internet traffic. To achieve that we use an iptables rule to
masquerade (aka SNAT - to make it seem as if packets came from the
itself) traffic that is bound for IPs outside the GCE project network
iptables -t nat -A POSTROUTING ! -d 10.0.0.0/8 -o eth0 -j MASQUERADE
Lastly we enable IP forwarding in the kernel (so the kernel will process packets for bridged containers):
The result of all this is that all
Pods can reach each other and can egress
traffic to the internet.
If you have a “dumb” L2 network, such as a simple switch in a “bare-metal” environment, you should be able to do something similar to the above GCE setup. Note that these instructions have only been tried very casually - it seems to work, but has not been thoroughly tested. If you use this technique and perfect the process, please let us know.
Follow the “With Linux Bridge devices” section of this very nice tutorial from Lars Kellogg-Stedman.
Weave Net is a
resilient and simple to use network for Kubernetes and its hosted applications.
Weave Net runs as a CNI plug-in
or stand-alone. In either version, it doesn’t require any configuration or extra code to run, and in both cases, the network provides one IP address per pod - as is standard for Kubernetes.
Flannel is a very simple overlay network that satisfies the Kubernetes requirements. It installs in minutes and should get you up and running if the above techniques are not working. Many people have reported success with Flannel and Kubernetes.
OpenVSwitch is a somewhat more mature but also complicated way to build an overlay network. This is endorsed by several of the “Big Shops” for networking.
Calico uses BGP to enable real container IPs.
Romana is an open source software defined networking (SDN) solution that lets you deploy Kubernetes without an overlay network.
The early design of the networking model and its rationale, and some future plans are described in more detail in the networking design document.