Anvil Deployment

This page documents known working practices for running Floability on Anvil.

First, follow the basic flow in Run your first backpack.

Run from an Anvil login node:

floability run --backpack <backpack-root> --batch-type slurm

Notebook/Jupyter runs on the login node. Distributed tasks are submitted through vine_factory to Slurm workers.

Anvil uses Slurm, so use --batch-type slurm.

Manager Ports

We could not find a published Anvil document that clearly lists which manager communication ports are open. In our tests, the following range worked:

floability run --backpack <backpack-root> --batch-type slurm --manager-ports 35000,40000

Data Location

We recommend storing large data/cache content in scratch space.

Example:

floability run --backpack <backpack-root> --batch-type slurm \
    --data-cache-dir /anvil/scratch/<username>/floability-cache-base-dir

Slurm Partition (Queue) Selection

On some jobs, you may need to specify a Slurm partition explicitly. Use --batch-options to pass scheduler options to vine_factory.

Example:

floability run --backpack <backpack-root> --batch-type slurm \
    --batch-options "-p wide"

For other partition/queue options, see: https://www.rcac.purdue.edu/knowledge/anvil/run/partitions

Checklist

  • Activate environment on the login node.
  • Confirm Slurm queue/account settings.
  • Add site-specific --batch-options if required.

If It Fails

See Deployment Overview and Troubleshooting.