Write plugins in Python

    Install

    To install the plugin server and PDK globally, use :

    A Kong Gateway Python plugin implementation has following attributes:

    1. Schema = (
    2.     { "message": { "type": "string" } },
    3. )
    4. version = '0.1.0'
    5. class Plugin(object):
    6.   pass
    • A class named Plugin defines the class that implements this plugin.
    • A dictionary called Schema that defines expected values and data types of the plugin.
    • The variables version and priority that define the version number and priority of execution respectively.

    Phase handlers

    You can implement custom logic during the following phases using the same function signature:

    • certificate
    • access
    • response
    • preread
    • log

    The presence of the response handler automatically enables the buffered proxy mode.

    Support for is available. To use type hints and autocomplete in an IDE, add the kong parameter to the phase handler function:

    1. import kong_pdk.pdk.kong as kong
    2. class Plugin(object):
    3.         self.config = config
    4.     def access(self, kong: kong.kong):
    5.         host, err = kong.request.get_header("host")

    The first argument to start_dedicated_server defines the plugin name and must be unique.

    Example configuration

    To load plugins using the configuration file, you have to map existing Kong Gateway properties to aspects of your plugin. Below are two examples of loading plugins within kong.conf.

    1. pluginserver_names = my-plugin,other-one
    2. pluginserver_my_plugin_socket = /usr/local/kong/my-plugin.socket
    3. pluginserver_my_plugin_start_cmd = /path/to/my-plugin.py
    4. pluginserver_my_plugin_query_cmd = /path/to/my-plugin.py --dump
    5. pluginserver_other_one_socket = /usr/local/kong/other-one.socket
    6. pluginserver_other_one_start_cmd = /path/to/other-one.py
    7. pluginserver_other_one_query_cmd = /path/to/other-one.py -dump

    The socket and start command settings coincide with their defaults and can be omitted:

    If your workload is IO intensive, you can use the Gevent model by passing the -g flag to start_cmd in kong.conf. If your workload is CPU intensive, consider the multi-processing model by passing the -m flag to start_cmd in .

    More Information