The ng-cdi intent deployer is a service for translating intents into configuration changes to a network.

We currently support three main methods of declaring intents:

• Dialogflow: Intents are sourced from the dialogflow NLP platform. Dialogflow submits intents to the service using a webhook.
• NILE: NILE intents are submitted to the service using the /deploy route.
• REST: The service generates api routes for each intent we can handle.

You can view the api documentation at /docs

ngcdi-intent-tests.zip contains the dialogflow agent. Once you have a dialogflow agent set up, you can import the intents from there.

For connecting dialogflow and the intent deployer, I recommend using ngrok or similar.

Using ngrok: ngrok http <port>

Then place the printed url into the fulfillment/webhook section of dialogflow in the format: <url>/dialogflow_webook and fill in the 'basic auth' parameters with the values from WEBHOOK_USER and WEBHOOK_PASS respectively.

You also need to set up a telegram bot and set the token in the local env file and in dialogflow under integrations.

Make sure that 'Enable webhook call for this intent' is enabled on the intents in dialogflow.

You need a running ONOS instance for the intent deployer to work with, in the case the ONOS instance is running on a remote server, you can use the following SSH command to set up port forwarding:

ssh -N -L 8181:localhost:8181 -L 5000:localhost:5000 \
  -L 172.17.0.1:8181:localhost:8181 -L 172.17.0.1:5000:localhost:5000 \
  <onos server>

The 127.17.0.1 binding here is to expose the service inside the docker network.

The following env vars must be set (just generate a secure password):

WEBHOOK_PASS=<dialogflow/prometheus webhook http auth pass>

The following env vars can be set:

TELEGRAM_TOKEN=<telegram token for status updates>
ONOS_API_URL=<url of the onos api, defaults to https://onos.demo.ng-cdi.com/onos/v1/>
NGCDI_URL=<url of the 'reroute' api, defaults to https://api.demo.ng-cdi.com/api/>

For a local instance of https://github.com/ng-cdi/demo-intents, you should probably have the following env vars:

export ONOS_API_URL=http://localhost:8181/onos/v1/
export NGCDI_URL=http://localhost:5000/api

To run the intent deployer, you can run poetry run uvicorn app:app You can get poetry from https://python-poetry.org/

The compose file is set up and just needs the environment section filling in.

You'll need the port forwarding set up and exposed inside the docker network. YMMV with this on non linux machines.

Run with docker-compose up

• alerts/: Contains definitions of alert handlers, these are triggered by prometheus alertmanager.
• batfish/: Contains the batfish topology.
• intent_deployer/: Internal stuff, refer to each file for details.
• intents/: Contains definitions of each intent.
• neat_integration/: Code and models for testing configuration changes using NEAT.
• webhook/: Types and models relevant to Dialogflow.

The application automatically collects intents defined inside of intents/, python decorators are used to register these.

• @register_nile_handler(<name>) declares a function to extract relevant information from a NILE intent.
• @register_dialogflow_handler(<name>) declares a function to extract relevant information from a dialogflow intent. This is passed a python dictionary with string keys (dict[str, Any]). The name parameter to the decorator should be the same as the corresponding dialogflow intent name.

The callbacks for the above decorators should return an object describing the intent, a good way of doing this is using a pydantic object like so:

from pydantic import BaseModel

class ForwardIntent(BaseModel):
    src: str
    dst: str
    path: list[str]

ForwardIntent(src="h1", dst="h2", path=["s1", "s2"])

The intent executor is declared to handle the intent parsed by the nile handler/ dialogflow handler functions:

@register_intent_executor(<intent type>)
async def handle_intent(ctx: Context, intent: <intent type>):
    # ...

The Context type here contains information and methods to assist with the interactivity of the intent deployment. Context has a single method currently: confirm_if_needed(msg, callback), in a dialogflow context this is used to display a message and await confirmation from the user, in any other context the callback is executed immediately.

In a dialogflow context the passed Context will be an instance of DialogflowContext which contains helper methods for building the initial response, waiting for confirmation, and .get_notifier which returns a Notifier object, which can be used to send followup and unprompted messages.

The .confirm_if_needed method of Context objects can be used to confirm with the user if something should happen. Currently this only does anything with dialogflow contexts, for NILE/REST contexts we assume confirmation and execute the callback immediately.

The .confirm_if_needed method can only be used for the first confirmation, if you need any more you should test if the context is a dialogflow context then fetch the notifier to send another message. You can use the .wait_for_confirmation method as many times as needed.

An example of using the notifier to send followup messages:

if isinstance(ctx, DialogflowContext):
    notifier = ctx.get_notifier()
    if notifier is not None:
        await notifier.send("Hello")