pipelines/backend/api

Kubeflow Pipelines API

Before You Start

Tools needed:

• Docker
• Make

Set environment variable API_VERSION to the version that you want to generate. We use v1beta1 as example here.

export API_VERSION="v1beta1"

Auto-generation of Go client and swagger definitions

Use make generate command to generate clients using a pre-built api-generator image:

make generate

Code will be generated in:

• ./${API_VERSION}/go_client
• ./${API_VERSION}/go_http_client
• ./${API_VERSION}/swagger

Note: ./${API_VERSION}/swagger/pipeline.upload.swagger.json is manually created, while the rest ./${API_VERSION}/swagger/*.swagger.json are auto generated.

Auto-generation of Python client

This will generate the Python client for the API version specified in the environment variable.

./build_kfp_server_api_python_package.sh

Code will be generated in ./${API_VERSION}/python_http_client.

Auto-generation of API reference documentation

This directory contains API definitions. They are used to generate the API reference on kubeflow.org.

• Use the tools bootprint-openapi and html-inline to generate the API reference from kfp_api_single_file.swagger.json. These instructions have shown how to generate a single self-contained html file which is the API reference, from a json file.

• Use the above generated html to replace the html section, which is below the title section, in the file kubeflow-pipeline-api-spec.html

Note: whenever the API definition changes (i.e., the file kfp_api_single_file.swagger.json changes), the API reference needs to be updated.

Auto-generation of api generator image

make push

When you update the Dockerfile, to make sure others are using the same image as you do:

1. push a new version of the api generator image to gcr.io/ml-pipeline-test/api-generator:latest.
2. update the PREBUILT_REMOTE_IMAGE var in Makefile to point to your new image.
3. push a new version of the release tools image to gcr.io/ml-pipeline-test/release:latest, run make push in test/release/Makefile.