Skip to content

Using Spack for Package Management

Spack is software that is pre-configured in HUIT Open OnDemand to enable the easy installation and management of high performance compute applications. There is a wide variety of software available, and for packages with existing spack configurations, the installation is usually just a one line command from the terminal.

Managing a Spack installation will require heavy use of the terminal; there is no graphical interface available. If you need help with your course's environment, please reach out to atg@fas.harvard.edu.

However, since installations can take a while, and since you likely want a consistent environment, there are some best practices for managing a spack installation. You can manage any spack installation that you have access to, whether it's a Spack installation in a course shared folder for a course that you're on the teaching staff for, or a personal home directory spack installation.

Spack Environments

While you can just install packages directly to the Spack installations, Spack environments provide more stability, especially when being loaded automatically. For example, if you install one version of a package, load it with a spack load command in your .bashrc file, and later install another version of the same package, then the spack load command in your .bashrc can suddenly stop working. Environments have an internal lock file that articulates the exact packages that are loaded, so they're safe to load from a .bashrc file or in an automated script. When Spack is required for an interactive application, we use Spack environments for this reason.

The Spack documentation provides a tutorial on environments, which is useful for a more comprehensive understanding of how these environments function. In this documentation, we'll take a more focused approach, dealing with common setup and installation tasks for preparing an environment for a course.

You can only set up environments in environments that you can modify. In HUIT Open OnDemand, that means that you can only set up environments in your own personal Spack installation in your home folder, or in a shared course folder. In both cases, you should have the global spack installation set as an upstream, whether by the course shared folder's pre-made spack installation script, or by following our directions on setting up your own Spack installation.

For the rest of this document, we'll assume that you've already sourced the setup-env.sh script from the installation where you want the environment to be.

Creating & Activating

For an environment to be used, it must first be created. This is pretty straightforward, with the command

spack env create environmentName

Replace "environmentName" with whatever you want the environment to be called. Then the environment can be activated with

spack env activate environmentName

Adding Packages

To get started, you can add one or more packages. You do this with a command like

spack add packageSpec

You can install a package simply by name, but if you have specific version or compiler requirements, you can specify those with Spack's spec syntax.

Adding a package won't install it, but it will associate that package with the environment itself.

Installing Packages

The syntax to start an installation is simple, but where to run the installation is less simple. It's best to run these installations in a Slurm job, since some installations will take a long time to run. Therefore, we recommend adapting this bash script for your Spack installation, and using it via sbatch. This is the script that we use to manage the global installation. You can adapt it for your use by copying the script to a file in your own home directory (we call it spackEnvironment.sh, so you might as well use the same name). Once you have your own copy of the script in HUIT OOD, you can edit the line marked with # CHANGEME to point the script to your own Spack installation, rather than the global one.

#!/bin/bash
if [ -x "${1}" ]
  then
    echo "Environment name is required, e.g. `spackEnvironment.sh environmentName`"
    exit 1
fi

# Change the line below to activate your own Spack environment
. /shared/spack/share/spack/setup-env.sh # CHANGEME
ENVIRONMENT=$1
echo "Activating environment $ENVIRONMENT..."
spack env activate $ENVIRONMENT
echo "Installing packages for $ENVIRONMENT..."
spack install
echo "Done!"

From wherever you place that script, assuming you name it spackEnvironment.sh, you can run it as a Slurm job with

sbatch -c 4 spackEnvironment.sh environmentName

Once you've submitted the job, you can check on the job's progress with slurm commands. For instance, you can see the state of your running jobs with the command squeue -u $(whoami). A "state" of "CF" indicates your job is allocated to a job that is currently starting (or configuring), and a state of "R" indicates that your job is running. This command will also show you the job ID of your job, which you can use to view the output of your job. The output file will appear in the directory you started the job from as slurm-$JOBID.out, where $JOBID is your job ID.

If you prefer to do this interactively, you can start an interactive Slurm session with an salloc command, then using srun spack install. However, this could require babysitting your terminal to see if there are any errors with the installation. By running the installation as a batch job, output from the install process is recorded to an output log file, so you can review what the script did at your leisure.

Saving an Environment Definition

If you have spent some time creating a Spack environment during the course of a term, you may want to re-use the same environment in future terms. Or, if you leave the university, you may want to save an environment definition to take with you to another institution to use on their infrastructure. Spack supports this with spack.yaml files.

When you create an environment, a file is created at $SPACK_ROOT/var/spack/environments/$ENVIRONMENT_NAME/spack.yaml, where $SPACK_ROOT is the location of your spack installation, and $ENVIRONMENT_NAME is the name of the environment that you've set up. You can copy this file to another location where you can download it. Say you save it as MyEnvironment.yaml. Then, in the future, you can re-create the same environment with 

spack env create environmentName MyEnvironment.yaml