Quick Start for Developers

Unix Shell

Currently, polaris only supports bash and related unix shells (such as ksh on the Mac). We do not support csh, tcsh or other variants of csh. An activation script for those shells will not be created.

If you normally use csh, tcsh or similar, you will need to temporarily switch to bash by calling /bin/bash each time you want to use polaris.

Set up a polaris repository: for beginners

To begin, obtain the main branch of the polaris repository with:

git clone git@github.com:E3SM-Project/polaris.git
cd polaris
git submodule update --init --recursive

There are 3 submodules with different versions of E3SM (E3SM-Project for MPAS-Ocean, Omega for Omega and MALI-Dev for MALI) in a e3sm_submodules directory of the polaris repository.

polaris conda environment, spack environment, compilers and system modules

As a developer, you will need your own conda environment with the latest dependencies for polaris and a development installation of polaris from the branch you’re working on. On supported machines, you will also need to point to a shared spack environment with some tools and libraries built for that system that polaris needs.

In the root of the repository is a tool, configure_polaris_envs.py that can get you started.

You will need to run ./configure_polaris_envs.py each time you check out a new branch or create a new worktree with git. Typically, you will not need to run this command when you make changes to files within the polaris python package. These will automatically be recognized because polaris is installed into the conda environment in “editable” mode. You will need to run the command if you add new code files or data files to the package because these don’t get added automatically.

Whether you are on one of the Supported Machines or an “unknown” machine, you will need to specify a path where Miniforge3 either has already been installed or an empty directory where the script can install it. You must have write permission in the base environment (if it exists).

Note

We have found that an existing Miniconda3 installation does not always work well for polaris, so please start with Miniforge3 instead.

Note

It is very important that you not use a shared installation of Miniforge3 or Miniconda3 such as the base environment for E3SM-Unified for polaris development. Most developers will not have write access to shared environments, meaning that you will get write-permission errors when you try to update the base environment or create the polaris development environment.

For anyone who does have write permission to a shared environment, you would be creating your polaris development environment in a shared space, which could cause confusion.

Please use your own personal installation of Miniforge3 for development, letting configure_polaris_envs.py download and install Miniforge3 for you if you don’t already have it installed.

Supported machines

If you are on one of the Supported Machines, run:

./configure_polaris_envs.py --conda <base_path_to_install_or_update_conda> \
    [-c <compiler>] [--mpi <mpi>] [-m <machine>] [--with_albany] \
    [--with_netlib_lapack] [--with_petsc]

The <base_path_to_install_or_update_conda> is typically ~/miniforge3. This is the location where you would like to install Miniforge3 or where it is already installed. If you have limited space in your home directory, you may want to give another path. If you already have it installed, that path will be used to add (or update) the polaris test environment.

See the machine under Supported Machines for a list of available compilers to pass to -c. If you don’t supply a compiler, you will get the default one for that machine. Typically, you will want the default MPI flavor that polaris has defined for each compiler, so you should not need to specify which MPI version to use but you may do so with --mpi if you need to.

If you are on a login node, the script should automatically recognize what machine you are on. You can supply the machine name with -m <machine> if you run into trouble with the automatic recognition (e.g. if you’re setting up the environment on a compute node, which is not recommended).

Environments with Albany

If you are working with MALI, you should specify --with_albany. This will ensure that the Albany and Trilinos libraries are included among those built with system compilers and MPI libraries, a requirement for many MAlI test cases. Currently, only Albany is only supported with gnu compilers.

It is safe to add the --with_albany flag for MPAS-Ocean but it is not recommended unless a user wants to be able to run both models with the same conda/spack environment. The main downside is simply that unneeded libraries will be linked in to MPAS-Ocean.

Environments with PETSc and Netlib-LAPACK

If you are working with MPAS-Ocean tasks that need PETSC and Netlib-LAPACK, you should specify --with_petsc --with_netlib_lapack to point to Spack environments where these libraries are included. Appropriate environment variables for pointing to these libraries will be build into the resulting load script (see below).

Unknown machines

If you are on an “unknown” machine, typically a Mac or Linux laptop or workstation, you will need to specify which flavor of MPI you want to use (mpich or openmpi):

./configure_polaris_envs.py --conda <conda_path> --mpi <mpi>

Again, the <conda_path> is typically ~/miniforge3, and is the location where you would like to install Miniforge3 or where it is already installed. If you already have it installed, that path will be used to add (or update) the polaris test environment.

We only support one set of compilers for Mac and Linux (gnu for Linux and clang with gfortran for Mac), so there is no need to specify them. See Other Machines for more details.

In addition, unknown machines require a config file to be specified when setting up the polaris test environment. A config file can be specified using -f <filename>, where <filename> is an absolute or relative path to the file. More information, including example config files, can be found in Config Files.

Note

Currently, there is not a good way to build Albany for an unknown machine as part of the polaris deployment process, meaning MALI will be limited to the shallow-ice approximation (SIA) solver.

To get started on HPC systems that aren’t supported by Polaris, get in touch with the developers.

What the script does

In addition to installing Miniforge3 and creating the conda environment for you, this script will also:

• install Jigsaw and Jigsaw-Python from source from the jigsaw-python submodule. These tools are used to create many of the meshes used in Polaris.

• install the polaris package from the local branch in “development” mode so changes you make to the repo are immediately reflected in the conda environment.

• with the --update_spack flag on supported machines, installs or reinstalls a spack environment with various system libraries. The --spack flag can be used to point to a location for the spack repo to be checked out. Without this flag, a default location is used. Spack is used to build several libraries with system compilers and MPI library, including: SCORPIO (parallel i/o for E3SM components) ESMF (making mapping files in parallel), MOAB, Trilinos, Albany, Netlib-LAPACK and PETSc. Please uses these flags with caution, as they can affect shared environments! See Deploying a new spack environment.

• with the --with_albany flag, creates or uses an existing Spack environment that includes Albany and Trilinos.

• with the --with_petsc --with_netlib_lapack flags, creates or uses an existing Spack environment that includes PETSc and Netlib-LAPACK.

• make an activation script called load_*.sh, where the details of the name encode the conda environment name, the machine, compilers, MPI libraries, and optional libraries, e.g. load_dev_polaris_<version>_<machine>_<compiler>_<mpi>.sh (<version> is the polaris version, <machine> is the name of the machine, <compiler> is the compiler name, and mpi is the MPI flavor).

• optionally (with the --check flag), run some tests to make sure some of the expected packages are available.

Optional flags

--check

Check to make sure expected commands are present

--python

Select a particular python version (the default is currently 3.8)

--env_name

Set the name of the environment (and the prefix for the activation script) to something other than the default (dev_polaris_<version> or dev_polaris_<version>_<mpi>).

--update_jigsaw

Used to reinstall Jigsaw and Jigsaw-Python into the conda environment if you have made changes to the Jigsaw (c++) code in the jigsaw-python submodule. You should not need to reinstall Jigsaw-Python if you have made changes only to the python code in jigsaw-python, as the python package is installed in edit mode.

Activating the environment

Each time you want to work with polaris, you will need to run:

source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh

This will load the appropriate conda environment, load system modules for compilers, MPI and libraries needed to build and run E3SM components, and set environment variables needed for E3SM components or polaris. It will also set an environment variable LOAD_POLARIS_ENV that points to the activation script. Polaris uses this to make an symlink to the activation script called load_polaris_env.sh in the work directory. When the load script is executed from the base of the polaris repository (i.e., as source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh), it will install the version of the polaris package from that location into the associated conda environment. When the load script is executed from the work directory through the symlink, it will activate the associated conda environment, but does not install the polaris package into the conda environment; it is assumed that is already up to date from when the conda environment was created or last updated.

It is generally recommended to activate the polaris environment (from either the polaris repo or via the workdir symlink) from a clean environment. Unexpected behavior may occur if activating a different polaris environment after having one already activated.

If you switch between different polaris branches, it is safest to rerun ./configure_polaris_envs.py with the same arguments as above to make sure dependencies are up to date and the polaris package points to the current directory. If you are certain that no polaris dependencies are different between branches, you can also simply source the activation script (load_*.sh) in the branch.

Once you have sourced the activation script, you can run polaris commands anywhere, and it always refers to that branch. To find out which branch you are actually running polaris from, you should run:

echo $LOAD_POLARIS_ENV

This will give you the path to the load script, which will also tell you where the branch is. If you do not use the worktree approach, you will also need to check what branch you are currently on with git log, git branch or a similar command.

If you wish to work with another compiler, simply rerun the script with a new compiler name and an activation script will be produced. You can then source either activation script to get the same conda environment but with different compilers and related modules. Make sure you are careful to set up polaris by pointing to a version of the MPAS model that was compiled with the correct compiler.

Switching between different polaris environments

Many developers are switching between different polaris branches. We have 2 main workflows for doing this: checking out different branches in the same directory (with git checkout) or creating new directories for each branch (with git worktree). Either way, you need to be careful that the version of the polaris package that is installed in the conda environment you are using is the one you want. But how to handle it differs slightly between these workflows.

If you are developing or using multiple polaris branches in the same directory (switching between them using git checkout), you will need to make sure you update your polaris environment after changing branches. Often the branches you’re developing will make use of the same conda environment, because they are using the same polaris version (so the dependencies aren’t changing). The same conda environment (e.g. dev_polaris_<version>) can safely be used with multiple branches if you explicitly reinstall the polaris package you want to use into the conda environment after moving to a new branch. You can do this by simply re-executing source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh from the root of the repo before proceeding.

Similarly, if you are developing or using multiple polaris branches but you use a different directory for each (creating the directories with git worktree), you will need to make sure the version of the polaris package in your conda environment is the one you want. If your branches use the same polaris version (so the dependencies are the same), you can use the same conda environment (e.g. dev_polaris_<version>) for all of them. But you will only be able to test one of them at a time. You will tell the conda environment which branch to use by running source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh from the root of the directory (worktree) you want to work with before proceeding.

In both of these workflows, you can modify the polaris code and the conda environment will notice the changes as you make them. However, if you have added or removed any files during your development, you need to source the load script again: source ./load_<conda_env>_<machine>_<compiler>_<mpi>.sh in the root of the repo or worktree so that the added or removed files will be accounted for in the conda environment.

If you know that polaris has different dependencies in a branch or worktree you are working on compared to a previous branch you have worked with (or if you aren’t sure), it is safest to not just reinstall the polaris package but also to check the dependencies by re-running: ./configure_polaris_envs.py with the same arguments as above. This will also reinstall the polaris package from the current directory. The activation script includes a check to see if the version of polaris used to produce the load script is the same as the version of polaris in the current branch. If the two don’t match, an error like the following results and the environment is not activated:

$ source load_polaris_test_morpheus_gnu_openmpi.sh
This load script is for a different version of polaris:
__version__ = '0.2.0'

Your code is version:
__version__ = '0.3.0-alpha.1'

You need to run ./configure_polaris_envs.py to update your conda
environment and load script.

If you need more than one conda environment (e.g. because you are testing multiple branches at the same time), you can choose your own name for the conda environment. Typically, this might be something related to the name of the branch you are developing. This can be done with the --env_name argument to ./configure_polaris_envs.py. You can reuse the same custom-named environment across multiple branches if that is useful. Just remember to reinstall polaris each time you switch branches.

Note

If you switch branches and do not remember to recreate the conda environment (./configure_polaris_envs.py) or at least source the activation script (load_*.sh), you are likely to end up with an incorrect and possibly unusable polaris package in your conda environment.

In general, if one wishes to switch between environments created for different polaris branches or applications, the best practice is to end the current terminal session and start a new session with a clean environment before executing the other polaris load script. Similarly, if you want to run a job script that itself sources the load script, it’s best to start a new terminal without having sourced a load script at all.

Note

With the conda environment activated, you can switch branches and update just the polaris package with:

python -m pip install --no-deps --no-build-isolation -e .

The activation script will do this automatically when you source it in the root directory of your polaris branch. The activation script will also check if the current polaris version matches the one used to create the activation script, thus catching situations where the dependencies are out of date and the configure script needs to be rerun. Since sourcing the activation script is substantially faster than rerunning the configure script, it is best to try the activation script first and run the configure script only if you have to.

Troubleshooting

If you run into trouble with the environment or just want a clean start, you can run:

./configure_polaris_envs.py --conda <conda_path> -c <compiler> --recreate

The --recreate flag will delete the conda environment and create it from scratch. This takes just a little extra time.

Creating/updating only the polaris environment

For some workflows (e.g. for MALI development with the Albany library when the MALI build environment has been created outside of polaris, for example, on an unsupported machine), you may only want to create the conda environment and not build SCORPIO, ESMF or include any system modules or environment variables in your activation script. In such cases, run with the --env_only flag:

./configure_polaris_envs.py --conda <conda_path> --env_only ...

Each time you want to work with polaris, you will need to run:

source ./load_<env_name>.sh

This will load the appropriate conda environment for polaris. It will also set an environment variable LOAD_POLARIS_ENV that points to the activation script. Polaris uses this to make a symlink to the activation script called load_polaris_env.sh in the work directory.

If you switch to another branch, you will need to rerun:

./configure_polaris_envs.py --conda <conda_path> --env_only

to make sure dependencies are up to date and the polaris package points to the current directory.

Note

With the conda environment activated, you can switch branches and update just the polaris package with:

python -m pip install --no-deps --no-build-isolation -e .

This will be substantially faster than rerunning ./configure_polaris_envs.py ... but at the risk that dependencies are not up-to-date. Since dependencies change fairly rarely, this will usually be safe.

Building E3SM components

There are 3 E3SM repositories that are submodules within the polaris repository.

For MPAS-Ocean and MALI both, see the last column of the table in Supported Machines for the right <mpas_make_target> command for each machine and compiler.

MPAS-Ocean

To build MPAS-Ocean, you would typically run:

source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh
cd e3sm_submodules/E3SM-Project/components/mpas-ocean/
make <mpas_make_target>

MALI

MALI should typically be compiled with the Albany library that contains the first-order velocity solver. The Albany first-order velocity solver is the only velocity option that is scientifically validated. On supported machines and with compilers that support albany, you can run:

./configure_polaris_envs.py --with_albany ...

Then, you can build MALI:

source ./load_<env_name>_<machine>_<compiler>_<mpi>_albany.sh
cd e3sm_submodules/MALI-Dev/components/mpas-albany-landice
make ALBANY=true <mpas_make_target>

Omega

If you simply wish to run the CTests from Omega, you likely want to use the Omega CTest Utility.

Otherwise, to build Omega,

source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh
git submodule update --init e3sm_submodules/Omega
cd e3sm_submodules/Omega
git submodule update --init --recursive externals/YAKL externals/ekat \
    externals/scorpio cime
cd components/omega
mkdir build
cd build
cmake \
   -DOMEGA_BUILD_TYPE=Release \
   -DOMEGA_CIME_COMPILER=${POLARIS_COMPILER} \
   -DOMEGA_CIME_MACHINE=${POLARIS_MACHINE} \
   -DOMEGA_METIS_ROOT=${METIS_ROOT} \
   -DOMEGA_PARMETIS_ROOT=${PARMETIS_ROOT} \
   -DOMEGA_BUILD_TEST=ON \
   -Wno-dev \
   -S .. \
   -B .

./omega_build.sh

You can remove -DOMEGA_BUILD_TEST=ON to skip building CTests. You can change -DOMEGA_BUILD_TYPE=Release to -DOMEGA_BUILD_TYPE=Debug to build in debug mode.

You can alter the example above to build whichever Omega branch and in whatever location you like. If you build in a location other than e3sm_submodules/Omega/components/omega/build, you will need to point to the relative or absolute path using the -p flag when you call polaris setup or polaris suite.

To set up tasks and suites to use Omega, you need to supply --model=omega to polaris setup or polaris suite. Otherwise, it will default to MPAS-Ocean.

Running polaris from the repo

If you follow the procedure above, you can run polaris with the polaris command-line tool exactly like described in the Quick Start for Users and as detailed in Command-line interface.

To list tasks you need to run:

polaris list

The results will be the same as described in Setting up tasks, but the tasks will come from the local polaris directory.

To set up a task, you will run something like:

polaris setup -t ocean/global_ocean/QU240/mesh -m $MACHINE -w $WORKDIR -p $COMPONENT

where $MACHINE is an ES3M machine, $WORKDIR is the location where polaris tasks will be set up and $COMPONENT is the directory where the E3SM component executable has been compiled. See polaris setup for details.

To list available suites, you would run:

polaris list --suites

And you would set up a suite as follows:

polaris suite -c ocean -t nightly -m $MACHINE -w $WORKDIR -p $COMPONENT

When you want to run the code, go to the work directory (for the suite or test case), log onto a compute node (if on an HPC machine) and run:

source load_polaris_env.sh
polaris run

The first command will source the same activation script (load_<env_name>_<machine>_<compiler>_<mpi>.sh) that you used to set up the suite or task (load_polaris_env.sh is just a symlink to that activation script you sourced before setting up the suite or task).

Code style for polaris

Polaris complies with the coding conventions of PEP8. Rather than memorize all the guidelines, the easiest way to stay in compliance as a developer writing new code or modifying existing code is to use a PEP8 style checker. When you create a load script, we automatically install pre-commit, a tools that helps to enforce this standard by checking your code each time you make a commit. It will tell you about various types of problems it finds. Internally, it uses flake8 to check PEP8 compliance, isort to sort, check and format imports, flynt to change any format strings to f-strings, and mypy to check for consistent variable types. An example error might be:

example.py:77:1: E302 expected 2 blank lines, found 1

For this example, we would just add an additional blank line after line 77 and try the commit again to make sure we’ve resolved the issue.

You may also find it useful to use an IDE with a PEP8 style checker built in, such as PyCharm. See this tutorial for some tips on checking code style in PyCharm.

Once you open a pull request for your feature, there is an additional PEP8 style check at this stage (again using pre-commit).

Set up a polaris repository with worktrees: for advanced users

This section uses git worktree, which provides more flexibility but is more complicated. See the beginner section above for the simpler version. In the worktree version, you will have many unix directories, and each corresponds to a git branch. It is easier to keep track of, and easier to work with many branches at once. Begin where you keep your repositories:

mkdir polaris
cd polaris
git clone git@github.com:E3SM-Project/polaris.git main
cd main

The E3SM-Project/polaris repo is now origin. You can add more remotes. For example:

git remote add mark-petersen git@github.com:mark-petersen/polaris.git
git fetch mark-petersen

To view all your remotes:

git remote -v

To view all available branches, both local and remote:

git branch -a

We will use the git worktree command to create a new local branch in its own unix directory:

cd polaris/main
git worktree add -b new_branch_name ../new_branch_name origin/main
cd ../new_branch_name

In this example, we branched off origin/main, but you could start from any branch, which is specified by the last git worktree argument.

There are two ways to build the E3SM component in standalone mode:

1. Submodules within polaris (easier): This guarantees that the E3SM commit that the submodule points to is compatible with polaris. It is also the default location for finding the E3SM component so you don’t need to specify the -p flag at the command line or put the E3SM component path path in your config file (if you even need a config file at all). Here is an example for MPAS-Ocean:

   source ./load_<env_name>_<machine>_<compiler>_<mpi>.sh
   git submodule update --init --recursive
   cd e3sm_submodules/E3SM-Project/components/mpas-ocean/
   make gfortran
   

2. Other E3SM directory (advanced): Create your own clone of the E3SM-Project/E3SM, E3SM-Project/Omega or MALI-Dev/E3SM repository elsewhere on disk. Either make a config file that specifies the absolute path to the path where the ocean_model or landice_model executable is found, or specify this path on the command line with -p. You are responsible for knowing if this particular version of MPAS component’s code is compatible with the version of polaris that you are using. The simplest way to set up a new repo for MALI development in a new directory is:

   git clone git@github.com:MALI-Dev/E3SM.git your_new_branch
   cd your_new_branch
   git checkout -b your_new_branch origin/develop
   

   The equivalent for MPAS-Ocean development would be:

   git clone git@github.com:E3SM-Project/E3SM.git your_new_branch
   cd your_new_branch
   git checkout -b your_new_branch origin/main