PyTorch-operator is currently at v1.
cd ${GOPATH}/src/github.com/kubeflow
git clone git@github.com:kubeflow/pytorch-operator.git
Resolve dependencies (if you don't have dep install, check how to do it here)
Install dependencies
dep ensure
Build it
go install github.com/kubeflow/pytorch-operator/cmd/pytorch-operator.v1
Running the operator locally (as opposed to deploying it on a K8s cluster) is convenient for debugging/development.
First, you need to run a Kubernetes cluster locally. There are lots of choices:
local-up-cluster.sh
runs a single-node Kubernetes cluster locally, but Minikube runs a single-node Kubernetes cluster inside a VM. It is all compilable with the controller, but the Kubernetes version should be 1.8
or above.
Notice: If you use local-up-cluster.sh
, please make sure that the kube-dns is up, see kubernetes/kubernetes#47739 for more details.
We can configure the operator to run locally using the configuration available in your kubeconfig to communicate with a K8s cluster. Set your environment:
export KUBECONFIG=$(echo ~/.kube/config)
export KUBEFLOW_NAMESPACE=$(your_namespace)
- KUBEFLOW_NAMESPACE is used when deployed on Kubernetes, we use this variable to create other resources (e.g. the resource lock) internal in the same namespace. It is optional, use
default
namespace if not set.
After the cluster is up, the PyTorch Operator CRD should be created on the cluster.
kubectl create -f ./manifests/crd.yaml
Now we are ready to run operator locally:
pytorch-operator.v1
To verify local operator is working, create an example job and you should see jobs created by it.
cd ./examples/mnist
docker build -f Dockerfile -t kubeflow/pytorch-dist-mnist-test:1.0 ./
kubectl create -f ./v1/pytorch_job_mnist_gloo.yaml
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.