Your AVD fabric, live in Kubernetes

Keep the AVD model you have and the containerlab you run. Change the model in Kubernetes, and the twin follows.

VS Code: a VLAN added to the AVD model with kubectl edit, beside the containerlab twin in TopoViewer and the Fabric's objects in the Kubernetes tree — A VLAN added to the model, in the cluster, beside its containerlab twin.

Crossview: the Fabric in the middle and a Device per switch around it, all healthy — The fabric in Kubernetes: one Fabric, a Device per switch.

show vlan on dc1-leaf1a listing the new VLAN 3403, beside the trace of the Fabric's objects, all ready — Half a minute later, the VLAN is on the switch.

What you get

The AVD model as it is: your inventory, group_vars and playbook become Kubernetes objects in AVD's own data model. No new schema.
Before anything moves, netadopt avd report names what cannot be carried, and netadopt avd compare renders the repository and the objects and compares every host.
A twin cabled as AVD cables it: in containerlab, or in Kubernetes with netclab-chart.
A change to the model reaches the switches in under a minute, with no playbook to run.
A change made by hand on a switch is put back within a minute.
The twin and production run from the same objects; only the lab's extra vars differ.

Quick start

AVD's own single-dc-l3ls example: 8 cEOS switches and 2 servers in containerlab, the model in a kind cluster. Run with netadopt 0.9.0, function-avd v0.2.4, Crossplane v2.4.2, containerlab 0.79.0 and cEOS 4.36.1F.

It needs Docker, kind, kubectl, helm, uv and containerlab. cEOS cannot be pulled: download it from Arista and import it once.

docker import cEOS64-lab-4.36.1F.tar.xz ceos:4.36.1F

1. The repository

git clone --depth 1 --branch v6.4.0 --filter=blob:none --sparse https://github.com/aristanetworks/avd
git -C avd sparse-checkout set ansible_collections/arista/avd/examples/single-dc-l3ls
cp -r avd/ansible_collections/arista/avd/examples/single-dc-l3ls .

2. What is in it

uvx "netadopt[avd]" avd report single-dc-l3ls --playbook build.yml

It carries: a Fabric, and a FabricInput per group.

3. The twin

uvx "netadopt[avd]" avd lab single-dc-l3ls --playbook build.yml --for containerlab \
  --ceos-image ceos:4.36.1F --extra-vars-out lab-vars.yml > single-dc-l3ls.clab.yml
sudo containerlab deploy -t single-dc-l3ls.clab.yml

lab-vars.yml points each switch of the model at its node in the twin. cEOS takes a few minutes to boot. In VS Code with vscode-containerlab, one setting logs into the nodes as the twin's user:

"containerlab.node.sshUserMapping": { "arista_ceos": "arista" }

4. The cluster

curl -LO https://raw.githubusercontent.com/netclab/function-avd/v0.2.4/example/runtime.yaml
curl -LO https://raw.githubusercontent.com/netclab/function-avd/v0.2.4/example/providerconfig.yaml
uvx netclab up --crossplane v2.4.2 \
  --configuration xpkg.upbound.io/netclab/configuration-avd:v0.2.4 \
  --manifest runtime.yaml --manifest providerconfig.yaml

A kind cluster with Crossplane and function-avd, in about a minute and a half.

5. The model in the cluster

uvx "netadopt[avd]" avd emit single-dc-l3ls --playbook build.yml -e @lab-vars.yml \
  | kubectl apply --server-side -f -
kubectl get fabric,device

The Fabric and its 8 Devices are ready in under a minute, and every switch in the twin carries its AVD configuration.

6. A change

KUBE_EDITOR="code --wait" kubectl edit fabricinput single-dc-l3ls-network-services

Under l2vlans, add:

- id: 3403
  name: NEW_VLAN

Save and close. Half a minute later the VLAN is on the six leaves, and not on the spines:

docker exec clab-single-dc-l3ls-dc1-leaf1a Cli -p 15 -c "show vlan"

7. Production

uvx "netadopt[avd]" avd emit single-dc-l3ls --playbook build.yml > production.yaml
uvx "netadopt[avd]" avd emit single-dc-l3ls --playbook build.yml -e @lab-vars.yml > twin.yaml
diff production.yaml twin.yaml

Only spec.extraVars differs. Applied without it, the Fabric configures the switches at the addresses in your inventory: read Before applying a Fabric first.

Clean up

sudo containerlab destroy -t single-dc-l3ls.clab.yml
uvx netclab down

What makes it possible

netadopt

Reads an AVD repository, says what it holds, and writes it as Kubernetes objects, and a twin of its switches.

PyPI Source
function-avd

Runs AVD in Kubernetes: builds each switch's configuration from those objects with AVD's own role, pushes it over eAPI, and pushes it again after a change by hand.

Source
Crossplane

The Kubernetes framework the objects are reconciled with.

crossplane.io
containerlab and netclab-chart

The twin: containerlab on your machine, or netclab-chart in the cluster.

containerlab netclab-chart