Set up a Kubernetes cluster for distributed AI research
It prepares a kubernetes cluster using terraform. It generates .tf.json files that are also recognized by Symphony.
- (Optional, Recommended) Create and work in a clean directory as running terraform would generate relevant files.
> mkdir surreal > cd surreal
You first need to setup credentials for terraform to access google cloud. See guide here. Choose one of the two methods:
- Run the following command
gcloud auth application-default login
- Go to the api key management page https://console.cloud.google.com/apis/credentials/serviceaccountkey and select Create new service account. You would need to give the service account sufficient permissions to do things properly. Project editor would suffice but is also more than enough. You can then generate and download the key, (json format is fine). Put the path to the .json file into the commandline argument when prompted.
Follow the instructions in the commandline tool.
It will provide instructions and generate a <cluster_name>.tf.json file which terraform recognizes. If you have generated a .json credential file, you should provide it when prompted. * terraform init && terraform plan describes changes to be made. * terraform apply makes the changes to your cloud project. * After cluster creation, obtain credentials for kubectl.
> gcloud container clusters get-credentials <cluster_name>
- If you have GPUs in your cluster, create the daemon set to install drivers, see documentation.
> kubectl apply -f https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/stable/nvidia-driver-installer/cos/daemonset-preloaded.yaml
- The generated <cluster_name>.tf.json is also recognized by Symphony’s scheduling mechanism and Surreal. So you may want to link to it
- If you want to remove everything, run terraform destroy
- Terraform install fails.
- If you are seeing error: ... API has not been used in project...: during terraform apply, go to the Kubernetes Engine tab and/or Compute Engine tab on your google cloud console to enable their APIs.
- GPU nodes are not scaling up.
- Check if the driver installation daemon set is running (see documentation).
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size cloudwise-0.1.1-py3-none-any.whl (9.0 kB)||File type Wheel||Python version py3||Upload date||Hashes View|
|Filename, size cloudwise-0.1.1.tar.gz (9.1 kB)||File type Source||Python version None||Upload date||Hashes View|
Hashes for cloudwise-0.1.1-py3-none-any.whl