You are browsing documentation for an outdated version. See the latest documentation here.

Plugin Development - Extending the Admin API

Notes:

• This chapter assumes that you have a relative knowledge of Lapis.
• The Admin API extensions are available only for HTTP plugins, not Stream plugins.

Kong can be configured using a REST interface referred to as the Admin API. Plugins can extend it by adding their own endpoints to accommodate custom entities or other personalized management needs. A typical example of this is the creation, retrieval, and deletion (commonly referred to as “CRUD operations”) of API keys.

The Admin API is a Lapis application, and Kong’s level of abstraction makes it easy for you to add endpoints.

Module

kong.plugins.<plugin_name>.api

Add endpoints to the Admin API

Kong will detect and load your endpoints if they are defined in a module named:

"kong.plugins.<plugin_name>.api"

This module is bound to return a table with one or more entries with the following structure:

{
  ["<path>"] = {
     schema = <schema>,
     methods = {
       before = function(self) ... end,
       on_error = function(self) ... end,
       GET = function(self) ... end,
       PUT = function(self) ... end,
       ...
     }
  },
  ...
}

Where:

• <path> should be a string representing a route like /users (See Lapis routes & URL Patterns) for details. Notice that the path can contain interpolation parameters, like /users/:users/new.
• <schema> is a schema definition. Schemas for core and custom plugin entities are available via kong.db.<entity>.schema. The schema is used to parse certain fields according to their types; for example if a field is marked as an integer, it will be parsed as such when it is passed to a function (by default form fields are all strings).
• The methods subtable contains functions, indexed by a string.
  • The before key is optional and can hold a function. If present, the function will be executed on every request that hits path, before any other function is invoked.
  • One or more functions can be indexed with HTTP method names, like GET or PUT. These functions will be executed when the appropriate HTTP method and path is matched. If a before function is present on the path, it will be executed first. Keep in mind that before functions can use kong.response.exit to finish early, effectively cancelling the “regular” http method function.
  • The on_error key is optional and can hold a function. If present, the function will be executed when the code from other functions (either from a before or a “http method”) throws an error. If not present, then Kong will use a default error handler to return the errors.

For example:

local endpoints = require "kong.api.endpoints"

local credentials_schema = kong.db.keyauth_credentials.schema
local consumers_schema = kong.db.consumers.schema

return {
  ["/consumers/:consumers/key-auth"] = {
    schema = credentials_schema,
    methods = {
      GET = endpoints.get_collection_endpoint(
              credentials_schema, consumers_schema, "consumer"),

      POST = endpoints.post_collection_endpoint(
              credentials_schema, consumers_schema, "consumer"),
    },
  },
}

This code will create two Admin API endpoints in /consumers/:consumers/key-auth, to obtain (GET) and create (POST) credentials associated to a given consumer. On this example the functions are provided by the kong.api.endpoints library.

The endpoints module currently contains the default implementation for the most usual CRUD operations used in Kong. This module provides you with helpers for any insert, retrieve, update or delete operations and performs the necessary DAO operations and replies with the appropriate HTTP status codes. It also provides you with functions to retrieve parameters from the path, such as an Service’s name or id, or a Consumer’s username or id.

If endpoints-provided are functions not enough, a regular Lua function can be used instead. From there you can use:

• Several functions provided by the endpoints module.
• All the functionality provided by the PDK
• The self parameter, which is the Lapis request object.
• And of course you can require any Lua modules if needed. Make sure they are compatible with OpenResty if you choose this route.

local endpoints = require "kong.api.endpoints"

local credentials_schema = kong.db.keyauth_credentials.schema
local consumers_schema = kong.db.consumers.schema

return {
  ["/consumers/:consumers/key-auth/:keyauth_credentials"] = {
    schema = credentials_schema,
    methods = {
      before = function(self, db, helpers)
        local consumer, _, err_t = endpoints.select_entity(self, db, consumers_schema)
        if err_t then
          return endpoints.handle_error(err_t)
        end
        if not consumer then
          return kong.response.exit(404, { message = "Not found" })
        end

        self.consumer = consumer

        if self.req.method ~= "PUT" then
          local cred, _, err_t = endpoints.select_entity(self, db, credentials_schema)
          if err_t then
            return endpoints.handle_error(err_t)
          end

          if not cred or cred.consumer.id ~= consumer.id then
            return kong.response.exit(404, { message = "Not found" })
          end
          self.keyauth_credential = cred
          self.params.keyauth_credentials = cred.id
        end
      end,
      GET  = endpoints.get_entity_endpoint(credentials_schema),
      PUT  = function(self, db, helpers)
        self.args.post.consumer = { id = self.consumer.id }
        return endpoints.put_entity_endpoint(credentials_schema)(self, db, helpers)
      end,
    },
  },
}

On the previous example, the /consumers/:consumers/key-auth/:keyauth_credentials path gets three functions:

• The before function is a custom Lua function which uses several endpoints-provided utilities (endpoints.handle_error) as well as PDK functions (kong.response.exit). It also populates self.consumer for the subsequent functions to use.
• The GET function is built entirely using endpoints. This is possible because the before has “prepared” things in advance, like self.consumer.
• The PUT function populates self.args.post.consumer before calling the endpoints-provided put_entity_endpoint function.