KGrid Python Runtime

A KGrid runtime for Knowledge Objects in a native python environment that connects to an activator using the proxy adapter.

Prerequisites:

• Python 3.8 or higher
• pip

Installation from the python package:

• Run python -m pip install kgrid-python-runtime to download the latest package
• Run python -m kgrid_python_runtime to start the runtime

Installation from an image:

• Download the latest image from docker hub: docker pull kgrid/kgrid-python-runtime

• Use the following command to run the image on Linux:

 docker run --network host kgrid/kgrid-python-runtime

Or

• Use the following command to run the image on Windows:

 docker run -it -p :5000:5000 -e KGRID_PROXY_ADAPTER_URL=http://host.docker.internal:8080 kgrid/kgrid-python-runtime

This starts the runtime pointed to an activator running on the same system at localhost:8080

Endpoints

The runtime exposes two endpoints which can be used to see the details of the runtime and what has been activated

GET /info

Displays details about the runtime such as the running version and status.

GET /endpoints

Displays a list of the activated endpoints in the engine.

Configuration:

Set these environment variables to customize your runtime's settings.

KGRID_PYTHON_ENV_URL

• The address of this environment that the activator will use to communicate with it.
• Default value: http://localhost

KGRID_PYTHON_ENV_PORT

• The port this environment is available on.
• Default value:5000

KGRID_PROXY_ADAPTER_URL

• The url of the adapter this runtime will communicate with

• Default value:http://localhost:8080

• By default, the python runtime will tell the Kgrid Activator that it is started at http://localhost:5000.

KGRID_PYTHON_CACHE_STRATEGY

• The caching strategy of this runtime. It can take three values: never, always, or use_checksum
  • never - existing objects from the activator are overwritten on every activation call.
  • always - existing objects stored in the runtime will never be re-downloaded from the activator and the local pyshelf and context.json files must be deleted and the runtime restarted for the objects to be replaced.
  • use_checksum - objects will look for a checksum in the deployment descriptor sent over during activation and only re-download the object if that checksum has changed.
• Default value: never

KGRID_PROXY_HEARTBEAT_INTERVAL

• The frequency (in seconds) at which the runtime will ping the activator and attempt to reconnect if the connection has been broken. Can be set to any value above 5 seconds or 0 to disable the heartbeat.
• Default value:30

DEBUG

• Changes the logging level to debug, takes a boolean true/false
• Default: false

Creating a python Knowledge Object:

Just like other knowledge objects, python objects have 4 basic parts: service.yaml, deployment.yaml, metadata.txt, and a payload that can be any number of python files.

The packaging spec for knowledge objects can be found here.

An example KO with naan of hello, a name of neighbor, api version of 1.0, and endpoint welcome, a Deployment Specification might look like this:

/welcome:
  post:
    artifact:
      - "src/hello.js"
    engine: "python"
    function: "main"
    entry: "src/hello.js"

Where function is the name of the main javascript entry function.

You would then execute this endpoint to see the code work:

POST <activator url>/<naan>/<name>/<api version>/<endpoint>

In this example: POST <activator url>/hello/neighbor/1.0/welcome

The Service Specification for this object would likewise then be

openapi: 3.0.2
info:
  version: '1.0'
  title: 'Hello neighbor'
  description: An example of simple Knowledge Object
  license:
    name: GNU General Public License v3 (GPL-3)
    url: >-
      https://tldrlegal.com/license/gnu-general-public-license-v3-(gpl-3)#fulltext
  contact:
    name: KGrid Team
    email: kgrid-developers@umich.edu
    url: 'http://kgrid.org'
servers:
  - url: /js/neighbor
    description: Hello world
tags:
  - name: KO Endpoints
    description: Hello world Endpoints
paths:
  /welcome:
    post:
      ... 

In the Service Specification the servers.url must match the naan and name of the object (/js/neighbor) and the path must match the path in Deployment Specification (/welcome). The service spec conforms to the swagger OpenAPI spec.

If your python package requires other python packages, simply specify them in a file called requirements.txt at the root of your object:

package-name=0.1.5
other-package-name=1.3.5
third-package-name=1.5.4

Right now the python runtime only supports requirements.txt using pip, see the requirements.txt documentation.

That's it! as long as the payload is written in valid python, and the object is built to the spec, you're ready to go. Example python objects can be found in the example collection or in the example-objects directory in this repository.

For Developers

To run the app:

Clone this project and set the environment variable: PYTHONPATH to the project root.

Example (Ubuntu): export PYTHONPATH=~/Projects/kgrid-python-runtime

Run python kgrid_python_runtime/app.py runserver from the top level of the project.

Important Notes

• Editing the cache directly from the runtime's shelf will not propagate changes to the endpoints in the runtime. New KOs must come from the activator.

• The runtime will attempt to load any Knowledge Objects that were previously loaded onto its shelf before registering with the activator and acquiring its objects. The shelf directory can be deleted if there is a need to get all objects fresh from the activator.