Writing Profiles

As it’s been already mentioned, a profile is a bunch of YAML files that describe how to map native configuration and how to translate an object into native configuration. In order to read native configuration we will use parsers, to translate a YANG model into native configuration we will use translators.

Both parsers and translators follow three basic rules:

  1. One directory per module.
  2. One file per model.
  3. Exact same representation of the model inside the file:

For example:

$ tree napalm_yang/mappings/eos/parsers/config
├── napalm-if-ip
│   └── secondary.yaml
├── openconfig-if-ip
│   └── ipv4.yaml
├── openconfig-interfaces
│   └── interfaces.yaml
└── openconfig-vlan
    ├── routed-vlan.yaml
    └── vlan.yaml

4 directories, 5 files
$ cat napalm_yang/mappings/eos/parsers/config/openconfig-vlan/vlan.yaml
    (trimmed for brevity)

    (trimmed for brevity)
        (trimmed for brevity)
            (trimmed for brevity)

If we check the content of the file vlan.yaml we can clearly see two parts:

  • metadata - This part specifies what parser or translator we want to use as there are several depending on the type of data we are parsing from or translating to and some options that the parser/translator might need. For example:

        processor: XMLParser
            - method: _rpc
                  get: "<get-configuration/>"

In this case we are using the XMLParser parser and in order to get the data we need from the device we have to call the method _rpc with the args parameters. This is, by the way, an RPC call for a junos device.

  • vlan - This is the part that follows the model specification. In this case is vlan but in others it might be interfaces, addressess or something else, this will be model dependent but it’s basically whatever it’s not metadata. This part will follow the model specification and add rules on each attribute to tell the parser/translator what needs to be done. For example:

        _process: unnecessary
            _process: unnecessary
                    mode: xpath
                    xpath: "vlan-id"
                    from: "{{ parse_bookmarks['parent'] }}"

As we are dealing with a parser we have to specify the _process attribute at each step (translators require the attribute _process). There are two special types of actions; unnecessary and not_implemented. Both do exactly the same, skip any action and move onto the next attribute. The only difference is purely aesthetically and for documentation purposes.

Something else worth noting is that each attribute inside _process/_process is evaluated as a jinja2 template so you can do variable substitutions, evaluations, etc...