This image is used when submitting Argo workflows to run TensorFlow/PAI TensorFlow jobs. To build this image, you should follow the below steps:

1. Go to SQLFlow root directory

   cd $(git rev-parse --show-toplevel)

2. Build the sqlflow:dev docker image as described here, only needed when you haven't done it. For short, you can just execute below command:

   docker build -t sqlflow:dev -f docker/dev/Dockerfile .

3. Run sqlflow:dev image to build SQLFlow project. This process will generate a directory called build under current directory which contains all the binaries we need.

   docker run --rm -it \
   -v $GOPATH:/root/go \
   -v $HOME/.m2:/root/.m2 \
   -v $HOME/.cache:/root/.cache \
   -v $PWD:/work -w /work \
   sqlflow:dev

4. Run below command to build the sqlflow:step docker image. This command will package previously built binaries into the result image.

   docker build -t sqlflow:step -f docker/step/Dockerfile .

To use it in SQLFlow, specify the environment variable export SQLFLOW_WORKFLOW_STEP_IMAGE=sqlflow:step when you start sqlflow server, or set the image in a Kubernetes config file, you can see an example here.