Kepler Metal CI

Running Kepler on Equinix and AWS Metal instances with GitHub Actions

Equinix Metal Runner

This action is used to demo how to deploy a GitHub self hosted runner on Equinix Metal, run some workload, and delete the runner.

Setup

• Create GitHub personal access token following this instruction
• Create a project with the Equinix Metal Project Action to create the self hosted runner
• Run some workload
• Delete the runner with the Equinix Metal Sweeper Action. Note, this action only deletes the server that serves the self hosted runner. It doesn't sweep other servers as the Equinix Metal Sweeper does.

Importing Prometheus Snapshot data from Validation Runs

To import the Prometheus data from the validation runs, follow these steps:

• Navigate to the directory containing the snapshot tar file, typically found in:

docs/validation/<date>/<validator_dir>/prometheus-snapshot

Extract the tar file.

• Ensure you have a local Prometheus instance set up. Identify the location of its data directory
• Copy the extracted snapshot contents into Prometheus data directory

cp -r <snapshot_directory>/* <prometheus_data_directory>

Workflows

• Validation with Standalone Kepler: Deploys Kepler on baremetal and virtual machine on Equinix. Validates the energy prediction of Kepler on virtual machine.
• Validation with Model Server: Deploys Kepler on baremetal and Kepler with Kepler Model Server on virtual machine on Equinix. Validates the energy prediction of Kepler with Model Server on Virtual Machine.
• Training and Validation e2e with Single Server: Provisions Equinix Server, Trains Models locally on server, Validates trained models with Model Server on Virtual Machine.
• Training and Validation e2e with Isolated Servers: Provisions Equinix Server, Trains Models locally on Server, Provisions a separate Equinix Server, Validates trained models with Model Server on Virtual Machine on new Server