Example using GraphQL mutations to configure a DHCP server:

This assumes that the http-api is running:

'set service https api'

One can configure an address on an interface, and configure the DHCP server
to run with that address as default router by requesting these 'mutations'
in the GraphQL playground:

mutation {
  createInterfaceEthernet (data: {interface: "eth1",
                                  address: "192.168.0.1/24",
                                  description: "BOB"}) {
    success
    errors
    data {
      address
    }
  }
}

mutation {
  createDhcpServer(data: {sharedNetworkName: "BOB",
                          subnet: "192.168.0.0/24",
                          defaultRouter: "192.168.0.1",
                          nameServer: "192.168.0.1",
                          domainName: "vyos.net",
                          lease: 86400,
                          range: 0,
                          start: "192.168.0.9",
                          stop: "192.168.0.254",
                          dnsForwardingAllowFrom: "192.168.0.0/24",
                          dnsForwardingCacheSize: 0,
                          dnsForwardingListenAddress: "192.168.0.1"}) {
    success
    errors
    data {
      defaultRouter
    }
  }
}

To save the configuration, use the following mutation:

mutation {
  saveConfigFile(data: {fileName: "/config/config.boot"}) {
    success
    errors
    data {
      fileName
    }
  }
}

N.B. fileName can be empty (fileName: "") or data can be empty (data: {}) to
save to /config/config.boot; to save to an alternative path, specify
fileName.

Similarly, using the same 'endpoint' (meaning the form of the request and
resolver; the actual enpoint for all GraphQL requests is
https://hostname/graphql), one can load an arbitrary config file from a
path.

mutation {
  loadConfigFile(data: {fileName: "/home/vyos/config.boot"}) {
    success
    errors
    data {
      fileName
    }
  }
}


The GraphQL playground will be found at:

https://{{ host_address }}/graphql

An equivalent curl command to the first example above would be:

curl -k 'https://192.168.100.168/graphql' -H 'Content-Type: application/json' --data-binary '{"query": "mutation {createInterfaceEthernet (data: {interface: \"eth1\", address: \"192.168.0.1/24\", description: \"BOB\"}) {success errors data {address}}}"}'

Note that the 'mutation' term is prefaced by 'query' in the curl command.

What's here:

services
├── api
│   └── graphql
│       ├── graphql
│       │   ├── directives.py
│       │   ├── __init__.py
│       │   ├── mutations.py
│       │   └── schema
│       │       ├── dhcp_server.graphql
│       │       ├── interface_ethernet.graphql
│       │       └── schema.graphql
│       ├── recipes
│       │   ├── dhcp_server.py
│       │   ├── __init__.py
│       │   ├── interface_ethernet.py
│       │   ├── recipe.py
│       │   └── templates
│       │       ├── dhcp_server.tmpl
│       │       └── interface_ethernet.tmpl
│       └── state.py
├── vyos-configd
├── vyos-hostsd
└── vyos-http-api-server

The GraphQL library that we are using, Ariadne, advertises itself as a
'schema-first' implementation: define the schema; define resolvers
(handlers) for declared Query and Mutation types (Subscription types are not
currently used).

In the current approach to a high-level API, we consider the
Jinja2-templated collection of configuration mode 'set'/'delete' commands as
the Ur-data; the GraphQL schema is produced from those files, located in
'api/graphql/recipes/templates'.

Resolvers for the schema Mutation fields are dynamically generated using a
'directive' added to the respective schema field. The directive,
'@generate', is handled by the class 'DataDirective' in
'api/graphql/graphql/directives.py', which calls the 'make_resolver' function in
'api/graphql/graphql/mutations.py'; the produced resolver calls the appropriate
wrapper in 'api/graphql/recipes', with base class doing the (overridable)
configuration steps of calling all defined 'set'/'delete' commands.

Integrating the above with vyos-http-api-server is ~10 lines of code.

What needs to be done:

• automate generation of schema and wrappers from templated configuration
commands

• investigate whether the subclassing provided by the named wrappers in
'api/graphql/recipes' is sufficient for use cases which need to modify data

• encapsulate the manipulation of 'canonical names' which transforms the
prefixed camel-case schema names to various snake-case file/function names

• consider mechanism for migration of templates: offline vs. on-the-fly

• define the naming convention for those schema fields that refer to
configuration mode parameters: e.g. how much of the path is needed as prefix
to uniquely define the term