Skip to content

Latest commit

 

History

History
105 lines (71 loc) · 2.89 KB

developer_guide.md

File metadata and controls

105 lines (71 loc) · 2.89 KB
Raw

Building the Operator

Create a symbolic link inside your GOPATH to the location you checked out the code

mkdir -p ${GOPATH}/src/github.com/tensorflow
ln -sf ${GIT_TRAINING} ${GOPATH}/src/github.com/tensorflow/k8s

Resolve dependencies (if you don't have glide install, check how to do it here)

glide install
rm -rf  vendor/k8s.io/apiextensions-apiserver/vendor
  • The rm is needed to remove the vendor directory of dependencies that also vendor dependencies as these produce conflicts with the versions vendored by mlkube

Build it

go install github.com/tensorflow/k8s/cmd/tf_operator

Building all the artifacts.

To build the following artifacts:

  • Docker image for the operator
  • Helm chart for deploying it

You can run

pip install -r py/requirements.txt
python -m py.release local --registry=${REGISTRY}
  • The docker image will be tagged into your registry
  • The helm chart will be created in ./bin

Running the Operator Locally

Running the operator locally (as opposed to deploying it on a K8s cluster) is convenient for debugging/development.

We can configure the operator to run locally using the configuration available in your kubeconfig to communicate with a K8s cluster. Set your environment:

export KUBECONFIG=$(echo ~/.kube/config)
export MY_POD_NAMESPACE=default
export MY_POD_NAME=my-pod
  • MY_POD_NAMESPACE is used because the CRD is namespace scoped and we use the namespace of the controller to set the corresponding namespace for the resource.
  • TODO(jlewi): Do we still need to set MY_POD_NAME? Why?

Make a copy of grpc_tensorflow_server.py and create a config file named controller_config_file.yaml:

cp grpc_tensorflow_server/grpc_tensorflow_server.py /tmp/grpc_tensorflow_server.py

cat > /tmp/controller_config_file.yaml << EOL
grpcServerFilePath: /tmp/grpc_tensorflow_server.py
EOL

Now we are ready to run operator locally:

tf_operator -controller_config_file=/tmp/controller_config_file.yaml

The command creates a CRD tfjobs and block watching for creation of the resource kind. To verify local operator is working, create an example job and you should see jobs created by it.

kubectl create -f https://raw.githubusercontent.com/tensorflow/k8s/master/examples/tf_job.yaml

Go version

On ubuntu the default go package appears to be gccgo-go which has problems see issue golang-go package is also really old so install from golang tarballs instead.

Code Style

Python

  • Use two spaces for indents in keeping with Python style

  • To autoformat code

    autopep8 -i --indent-size=2 path/to/module.py

  • To sort imports

    isort path/to/module.py