During this quick-start tutorial, we have been using a docker image polyaxon/polyaxon-quick-start, which was built using a Polyaxonfile as well.

Note: You can build docker images outside of Polyaxon, or use your own system.

Docker images

Using the job runtime, users can use Polyaxon to build docker images, there are a couple of build options and components provided.

For example the image we are using in this tutorial is based on this Polyaxonfile:

version: 1
kind: operation
name: build
    connection: docker-connection
    value: polyaxon/polyaxon-quick-start
  - dockerfile:
      image: "tensorflow/tensorflow:2.0.1-py3"
      - 'pip3 install --no-cache-dir -U polyaxon'
      langEnv: 'en_US.UTF-8'
hubRef: kaniko

This configuration is using a public component called Kaniko, it uses an initializer dockerfile which generates simple dockerfile.

Note: We could have created a dockerfile, and used a git initializer to clone the repo containing the dockerfile.

Since we do not want to create or modify the Kaniko component, we are using the runPatch section to add the init section to the job. The runPatch allows us to patch the component definition without having to rewrite it from scratch, in this case, we are generating a dockerfile. The dockerfile we are generating is a based on a tensorflow docker image and we are just installing polyaxon library and some extra dependencies to use with the tracking module.

The hubRef is the reference of the component we are going to run, in this case, it’s a Kaniko component for building the image.

We are also passing a parameter destination which is of type image, it defines the name of the image and the connection to use for pushing the image. The docker-connection is a connection that we configured to authenticate Kaniko to push images.