Topology definition
Containerlab builds labs based on the topology information that users pass to it. This topology information is expressed as a code contained in the topology definition file which structure is the prime focus of this document.
Topology definition components#
The topology definition file is a configuration file expressed in YAML and has a name pattern of *.clab.yml
1. In this document, we take a pre-packaged Nokia SR Linux and Arista cEOS lab and explain the topology definition structure using its definition file srlceos01.clab.yml which is pasted below:
name: srlceos01
topology:
nodes:
srl:
kind: srl
image: ghcr.io/nokia/srlinux
ceos:
kind: ceos
image: ceos:4.25.0F
links:
- endpoints: ["srl:e1-1", "ceos:eth1"]
Tip
Containerlab provides a JSON schema file for the topology file. The schema is used to live-validate user's input if a code editor supports this feature.
Additionally, the auto-generated schema documentation can be explored to understand the full scope of the configuration options containerlab provides.
This topology results in the two nodes being started up and interconnected with each other using a single point-po-point interface:
Let's touch on the key components of the topology definition file used in this example.
Name#
The topology must have a name associated with it. The name is used to distinct one topology from another, to allow multiple topologies to be deployed on the same host without clashes.
Its user's responsibility to give labs unique names if they plan to run multiple labs.
The name is a free-formed string, though it is better not to use dashes (-
) as they are used to separate lab names from node names.
When containerlab starts the containers, their names will be generated using the following pattern: clab-{{lab-name}}-{{node-name}}
. The lab name here is used to make the container's names unique between two different labs, even if the nodes are named the same.
Prefix#
It is possible to change the prefix that containerlab adds to node names. The prefix
parameter is in charge of that. It follows the below-mentioned logic:
- When
prefix
is not present in the topology file, the default prefix logic applies. Containers will be named asclab-<lab-name>-<node-name>
. - When
prefix
is set to some value, for example,myprefix
, this string is used instead ofclab
, and the resulting container name will be:myprefix-<lab-name>-<node-name>
. - When
prefix
is set to a magic value__lab-name
the resulting container name will not have theclab
prefix, but will keep the lab name:<lab-name>-<node-name>
. - When set to an empty string, the node names will not be prefixed at all. If your node is named
mynode
, you will get themynode
container in your system.
Warning
In the case of an empty prefix, you have to keep in mind that nodes need to be named uniquely across all labs.
Examples:
Note
Even when you change the prefix, the lab directory is still uniformly named using the clab-<lab-name>
pattern.
Topology#
The topology object inside the topology definition is the core element of the file. Under the topology
element you will find all the main building blocks of a topology such as nodes
, kinds
, defaults
and links
.
Nodes#
As with every other topology the nodes are in the center of things. With nodes we define which lab elements we want to run, in what configuration and flavor.
Let's zoom into the two nodes we have defined in our topology:
topology:
nodes:
srl: # this is a name of the 1st node
kind: srl
type: ixrd2
image: ghcr.io/nokia/srlinux
ceos: # this is a name of the 2nd node
kind: ceos
image: ceos:4.25.0F
We defined individual nodes under the topology.nodes
container. The name of the node is the key under which it is defined. Following the example, our two nodes are named srl
and ceos
respectively.
Each node can have multiple configuration properties which make containerlab quite a flexible tool. The srl
node in our example is defined with the a few node-specific properties:
Refer to the node configuration document to meet all other options a node can have.
Links#
Although it is totally fine to define a node without any links (like in this lab) most of the time we interconnect the nodes to make datapaths. One of containerlab purposes is to make the interconnection of nodes simple.
Links are defined under the topology.links
container in the following manner:
# nodes configuration omitted for clarity
topology:
nodes:
srl:
ceos:
links:
- endpoints: ["srl:e1-1", "ceos:eth1"]
- endpoints: ["srl:e1-2", "ceos:eth2"]
As you see, the topology.links
element is a list of individual links. The link itself is expressed as pair of endpoints
. This might sound complicated, lets use a graphical explanation:
As demonstrated on a diagram above, the links between the containers are the point-to-point links which are defined by a pair of interfaces. The link defined as:
will result in a creation of a p2p link between the node named srl
and its e1-1
interface and the node named ceos
and its eth1
interface. The p2p link is realized with a veth pair.
Kinds#
Kinds define the behavior and the nature of a node, it says if the node is a specific containerized Network OS, virtualized router or something else. We go into details of kinds in its own document section, so here we will discuss what happens when kinds
section appears in the topology definition:
topology:
kinds:
srl:
type: ixrd2
image: ghcr.io/nokia/srlinux
nodes:
srl1:
kind: srl
srl2:
kind: srl
srl3:
kind: srl
In the example above the topology.kinds
element has srl
kind referenced. With this, we set some values for the properties of the srl
kind. A configuration like that says that nodes of srl
kind will also inherit the properties (type, image) defined on the kind level.
Essentially, what kinds
section allows us to do is to shorten the lab definition in cases when we have a number of nodes of a same kind. All the nodes (srl1
, srl2
, srl3
) will have the same values for their type
and image
properties.
Consider how the topology would have looked like without setting the kinds
object:
topology:
nodes:
srl1:
kind: srl
type: ixrd2
image: ghcr.io/nokia/srlinux
srl2:
kind: srl
type: ixrd2
image: ghcr.io/nokia/srlinux
srl3:
kind: srl
type: ixrd2
image: ghcr.io/nokia/srlinux
A lot of unnecessary repetition which is eliminated when we set srl
kind properties on kind level.
Defaults#
kinds
set the values for the properties of a specific kind, whereas with the defaults
container it is possible to set values globally.
For example, to set the environment variable for all the nodes of a topology:
Now every node in this topology will have environment variable MYENV
set to VALUE
.
Generated topologies#
Advanced topic
To further simplify parametrization of the topology files, containerlab allows users to template the topology files using Go Template engine.
Using templating approach it is possible to create a lab template and instantiate different labs from it, by simply changing the variables in the variables file.
Standard Go templating language has been extended with the functions provided in docs.gomplate.ca project, which opens the doors to a very flexible topology generation workflows.
To help you get started, we created the following lab examples which demonstrate how topology templating can be used:
- Leaf-Spine topology with parametrized number of leaves/spines
- 5-stage Clos topology with parametrized number of pods and super-spines
-
if the filename has
.clab.yml
or-clab.yml
suffix, the YAML file will have autocompletion and linting support in VSCode editor. ↩