GMX K8S Tuner

This project deploys a tuning API for GROMACS using Ray Tune on Kubernetes.

Prerequisites

• Helm and kubectl must be installed and configured to interact with your Kubernetes cluster.
• Ray operator must be installed in your cluster to provide the RayCluster CRDs.
• Ensure your target namespace has sufficient resources: a minimum of 32 CPU requests and a GPU request (see helm/charts/gromacs-tuner/templates/raycluster.yaml).
• jq must be installed for processing JSON data.
• Update the API Docker image details in values.yaml (refer to api/Dockerfile) with your Harbor image information, including the correct image name and tag.

Deployment Steps

1. Configure values.yaml:

   • Specify the ingress.host for accessing the API and the ingress.tlsSecretName for TLS configuration.
   • Provide the complete Docker image.repository and image.tag for the API.

2. Create Authentication Secret:

   Create a Kubernetes secret named tuner-auth in your target namespace containing the user and password for API authentication.

   # Replace <namespace> with your target namespace (e.g., gromacs-tuner-ns)
   kubectl create secret generic tuner-auth \
     --namespace <namespace> \
     --from-literal=user=admin \
     --from-literal=password=your-strong-password

3. Deploy:

   cd helm/
   make install

4. Verify the Deployment:

   Check if the pods are running correctly:

   kubectl get pods -n gromacs-tuner

   Verify that the Ray cluster is properly deployed:

   kubectl get rayclusters -n gromacs-tuner

Uninstalling

To remove everything:

cd helm/
make uninstall

Basic Workflow

1. Submit a TPR File:

   Execute the run_submit.sh script to submit your GROMACS .tpr file (e.g., md.tpr) for tuning:

   ./run_submit.sh /path/to/md.tpr

   This script submits the specified .tpr file to the API. The API will return a JSON response containing a tuner_run_id.

   The API is protected using HTTP Basic Auth. You must provide the admin username and the password configured in the Kubernetes Secret when accessing any endpoint.

2. Poll Job Status:

   Use the poll_status.sh script to monitor the tuning process:

   ./poll_status.sh <tuner_run_id>

   Replace <tuner_run_id> with the UUID obtained from the submission response. This script will provide updates on the tuning job's status.