NOTE: Spack-CM was designed for earlier versions of spack (e.g., 0.16.0). Newer functionality renders it relatively useless; however, it can still be useful for those who are using older versions.
The Software Engineering Maintenance and Support (SEMS) operations and maintenance (O&M) team is proud to present the Spack Configuration Manager (spack-cm).
This tool leverages Spack to install compilers, utilities, and TPLs. It creates Spack environments for each project with customizations available based on the platform (or machine).
Spack-cm has several core requirements:
We recommend, as a best practice, that a building area directory be created:
$ cd ~/
$ mkdir tpl_building
$ cd tpl_building
To install and activate spack, run the following:
# Get a copy of Spack
$ git clone https://github.com/spack/spack.git
# Activate Spack
$ source spack/share/spack/setup-env.sh
Most machines have a Python 3.7+ version already available. Run the the following to confirm:
$ python3 --version
If the result returns anything Python 3.7+, this requirement is already
satisfied. If the result is an older Python or an error, you will
need to install Python either from source
or using a package manager (such as brew on Macintosh).
We recommend that all work be done in a virtual environment. This removes the need for administrator permissions and keeps your local Python environment clean.
To create and activate a virtual environment, run:
# Create the environment
$ python3 -m venv ~/tpl_building/venv
# Activate the environment
$ source ~/tpl_building/venv/bin/activate
Please read the venv documentation
for more details.
PyYAML is available through the Python package installer pip. If you
are the administrator on the machine, you will be able to run:
$ pip install pyyaml
To install, run:
$ cd ~/tpl_building
# Get a copy of the tool
$ git clone git@github.com:sandialabs/spack-cm.git
$ cd spack-cm
# Install the tool
$ python setup.py develop
This will install the command-line executable spack-cm that can be
run from anywhere.
There are two stages in the usage of this tool: setup and install.
The setup phase of the spack-cm tool creates the required
directory infrastructure for installation to run properly.
This step only needs to be run ONCE per project/machine combination,
after which, the changes can be committed and pushed back to the spack-cm
repository for future use.
To complete this step, follow the below guidance:
$ spack-cm setup -h
usage: spack-cm setup [-h] [-p PROJECT] [-m ALTHOSTNAME] [--spack SPACKVERSION]
Run set up routine for a new project/machine combination.
optional arguments:
-h, --help show this help message and exit
-p PROJECT, --project PROJECT
REQUIRED: Project for which to install TPLs (e.g., sems, pyomo, etc.).
-m ALTHOSTNAME, --machine ALTHOSTNAME
OPTIONAL: Designate an alternate platform name (i.e., not the hostname of the machine).
--spack SPACKVERSION OPTIONAL: Version of spack. Default: 0.16.2
The project option is required, whereas the following two are optional.
Upon running this command, it will create the file structure:
$ tree src/project/
src/project/
├── projectname
│ ├── projectname-manifest.yaml
│ ├── projectname-repo
│ │ ├── packages
│ │ └── repo.yaml
│ └── repos.yaml
$ tree src/platform/
src/platform/
├── machinename
│ ├── licenses
│ ├── compilers.yaml
│ ├── mirrors.yaml
│ └── packages.yaml
Under src/project, a directory named after your project will be created.
In addition, there will be a file projectname-manifest.yaml, which is covered
in the Required Customization section below.
Under src/platform, a directory named after your machine will be created.
In addition, several empty YAML files that can be utilized by spack are
created, which are covered in the Optional Customization section below.
The install phase of the spack-cm tool runs the installation
of your project's requested packages.
This step will need to be run anytime something needs to be installed.
IMPORTANT: Before you run this step, we advise that you module purge
or at least module unload any unnecessary modules. There are known issues
with certain environment variables set by loaded modules. You will
not want to unload any compilers that you plan to use
to build TPLs.
To complete this step, follow the below guidance:
$ spack-cm install -h
usage: spack-cm install [-h] [-p PROJECT] [-m ALTHOSTNAME] [-r ROOT_PATH] [-s STAGE] [--spack SPACKVERSION] [--install-spack-deps] [-d] [-e]
[--no-project-modules] [--add-machine-to-install-path] [--dry-run]
Run install routine for a project/machine combination.
optional arguments:
-h, --help show this help message and exit
-p PROJECT, --project PROJECT
REQUIRED: Project for which to install TPLs (e.g., sems, pyomo, etc.).
-m ALTHOSTNAME, --machine ALTHOSTNAME
OPTIONAL: Designate an alternate platform name (i.e., not the hostname of the machine).
-r ROOT_PATH, --root ROOT_PATH
REQUIRED: Root path in which to install TPLs (e.g. /project/sems, /project/pyomo, etc.).
-s STAGE, --stage STAGE
OPTIONAL: Select a single stage of the install to run. By default, all stages will run. Available choices: [base, compiler,
utility, tpl]
--spack SPACKVERSION OPTIONAL: Version of spack. Default: 0.16.2
--install-spack-deps OPTIONAL: Install spack system dependencies.
-d, --debug OPTIONAL: Enable "spack --debug" install mode.
-e, --external OPTIONAL: Allow spack to find and use system packages.
--no-project-modules OPTIONAL: Turn off use of project name in modulefile generation.
--add-machine-to-install-path
OPTIONAL: Add the machine name to the install path.
--dry-run OPTIONAL: Only do a dry-run of an install for trial or debug purposes without installing anything.
As with the previous step, project is required. The root_path option is also
required. This option specifies where the packages should be installed. Under
the root_path, the installation tree will look like:
root_path/
├── install
│ └── projectname
│ ├── base-packages
│ ├── compiler
│ ├── lmod
│ ├── tpl
│ └── utility
└── modulefiles
└── projectname
This allows multiple projects to install in the same root_path without
overwriting files.
The installation is based on the list found in your project's manifest file.
As mentioned in the Setup section, when a project area is created, a sample
projectname-manifest.yaml file is also created, the contents of which are:
# Compiler with which base packages will be installed.
# NOTE: System compilers are unreliable for newer versions of some
# compilers, such as gcc-4.8.5 and gcc-10.X. We recommend a newer compiler.
# If left empty, SPACK_CM_BASE_COMPILER will default to the system compiler.
SPACK_CM_BASE_COMPILER:
- ''
# USER WARNING: This list should not be modified unless by an advanced user.
# Basic, common dependencies which will be built only once.
SPACK_CM_BASE_PACKAGES:
- autoconf
- automake
- bzip2
- curl
- diffutils
- gmp
- xz
- zlib
- zstd
# Compilers to be installed.
SPACK_CM_COMPILERS:
- gcc@7.3.0
- gcc@10.1.0
# Compilers NOT to be installed, but against which TPLs should be built.
# (e.g., externally installed compilers)
SPACK_CM_EXTERNAL_COMPILERS:
- ''
# Compiler with which utilities will be built.
# NOTE: System compilers are notoriously unreliable for newer versions of some
# utilities, such as CMake. We recommend a newer compiler.
# If left empty, SPACK_CM_UTILITY_COMPILER will default to the system compiler.
SPACK_CM_UTILITY_COMPILER:
- gcc@7.3.0
# Utilities to be installed.
SPACK_CM_UTILITIES:
- cmake
- git
- ninja
- vvtest
# MPIs to be installed.
SPACK_CM_MPIS:
- openmpi@4.0.5
# MPIs NOT to be installed, but against which TPLs should be built.
# (e.g., externally installed MPIs)
SPACK_CM_EXTERNAL_MPIS:
- ''
# Cudas to be installed.
SPACK_CM_CUDAS:
- ''
# Cudas NOT to be installed, but against which TPLs should be built.
# (e.g., externally installed cudas)
SPACK_CM_EXTERNAL_CUDAS:
- ''
# TPLs to be installed.
SPACK_CM_TPLS:
- hdf5@1.10.6
- parallel-netcdf@1.9.0
- parmetis@4.0.3
- superlu-dist@5.4.0
- superlu@4.3
# Exclude these combinations of TPLs/compilers
SPACK_CM_EXCLUDE_COMBOS:
- ''
This file must be customized to match the needs of your project and/or your platform
(some of the defaults aren't appropriate for Mac, for example). All items are
expected to be present but can be empty strings (see SPACK_CM_CUDAS as an example).
Edit this file with your project's desired packages before running the
install phase.
Within the machinename directory, a licenses directory is created.
You may place an Intel license.lic file in this directory, which will
be used when installing Intel.
There are many YAML files that allow advanced customization of Spack which
are supplied in the machinename and projectname areas.
See the Spack configuration docs for more details.