Scenarios¶

The package offers several mechanisms for configuring the same device. Each one is a directory under scenarios/ holding manifests you can apply as they are.

scenario	mechanism	what it configures
`restconf`	OpenConfig over RESTCONF	the base config — BGP, interfaces, routing
`jsonrpc`	eAPI over JSON-RPC	settings OpenConfig does not model
`eapi`	raw EOS CLI	the same config, expressed as commands
`router`	the `Router` abstraction	one device from one resource
`fabric`	an AVD design	a whole network from one model

Pick one mechanism per device

restconf, eapi and router each become an owner of the same configuration. Applying two of them to one device means two things editing the same lines. jsonrpc adds settings the others do not reach, and can sit alongside restconf.

fabric is different in kind — it configures a whole network and runs on devices of its own.

Before you start¶

Every command below runs from a clone of the repository — the scenarios are manifests, not something the package installs:

git clone https://github.com/netclab/netclab-xp
cd netclab-xp

You also need devices, the package installed, and its prerequisites applied, and yq — removing a scenario filters the rendered manifests through it.

Applying a scenario¶

Every scenario is applied the same way; the directory name is the only thing that changes. Substitute any row of the table above for router here and in the teardown below.

kubectl kustomize --load-restrictor LoadRestrictionsNone scenarios/router \
  | kubectl apply -f -

The flag is needed because the overlays reach into examples/, so that every manifest exists in exactly one place. To use a different namespace, change namespace: in that scenario's kustomization.yaml — the Namespace object follows it.

Removing a scenario¶

Filter the Namespace out of the delete:

kubectl kustomize --load-restrictor LoadRestrictionsNone scenarios/router \
  | yq 'select(.kind != "Namespace")' \
  | kubectl delete -f -

Deleting the Namespace together with the resources inside it deadlocks. The namespace begins terminating, which forbids creating anything in it — and provider-http's cleanup path needs to create a tracking object there. The requests keep their finalizers, the namespace waits for the requests, and neither finishes without editing finalizers by hand.

Leaving the namespace behind is also the right outcome: it is yours, not the scenario's.

kubectl delete returns long before the device changes

The resources are gone in a fraction of a second, but the requests that talk to the device are cleaned up afterwards. Watch those, not the command:

kubectl -n netclab get requests.http.m.crossplane.io -w

Deleting resources strips the device¶

This is the part that surprises people, and it is the point of the package: what Crossplane wrote, Crossplane removes. After deleting a restconf scenario, the BGP instance is gone, ip routing is gone, the loopback does not exist, and the ethernet interface is back to bare.

fabric behaves the opposite way — it pushes a device's entire running configuration, so a teardown that reverted it would wipe the switch. It stops managing what it wrote and leaves it in place. See the fabric walkthrough.

Readiness lags the device by one poll

A resource commonly reports Ready=False for a poll interval after its requests are already Ready=True, often next to Responsive=False WatchCircuitOpen. It clears on its own. Wait rather than debug it.

Each scenario's page carries its own apply and teardown commands, so the two above are the general shape rather than something to adapt by hand.