Software modules

Videos

Videos of this topic may be available from one of our kickstart course playlists: 2023, 2022 Summer, 2022 February, 2021 Summer, 2021 February.

There are hundreds of people using every cluster. They all have different software needs, including conflicting versions required for different projects! How do we handle this without making a mess, or one person breaking the cluster for everyone?

This is actually a very hard, but solved within certain parameters, problem. Software installation and management takes up a huge amount of our time, but we try to make it easy for our users. Still, it can end up taking a lot of your effort as well.

Abstract

  • We use the standard Lmod module system, which makes more software available by adjusting environment variables like PATH.

  • module spider NAME searches for NAME.

  • module load NAME loads the module of that name.

  • See the Triton quick reference for a module command cheatsheet.

Local differences

Almost every site uses modules, and most use the same Lmod system we use here. But, the exact module names you can load will be different.

Introduction to modules

The answer is the standard “module” system Lmod. It allows us to have unlimited number of different software packages installed, and the user can select what they want. Modules include everything from compilers (+their required development files), libraries, and programs. If you need a program installed, we will put it in the module system.

In a system the size of Triton, it just isn’t possible to install all software by default for every user.

A module lets you adjust what software is available, and makes it easy to switch between different versions.

As an example, let’s inspect the anaconda module with module show anaconda:

$ module show anaconda
----------------------------------------------------------------------------
  /share/apps/anaconda-ci/fgci-centos7-anaconda/modules/anaconda/2023-01.lua:
----------------------------------------------------------------------------
whatis("Name : anaconda")
whatis("Version : 2023-01")
help([[This is an automatically created Anaconda installation.]])
prepend_path("PATH","/share/apps/anaconda-ci/fgci-centos7-anaconda/software/anaconda/2023-01/2eea7963/bin")
setenv("CONDA_PREFIX","/share/apps/anaconda-ci/fgci-centos7-anaconda/software/anaconda/2023-01/2eea7963")
setenv("GUROBI_HOME","/share/apps/anaconda-ci/fgci-centos7-anaconda/software/anaconda/2023-01/2eea7963")
setenv("GRB_LICENSE_FILE","/share/apps/manual_installations/gurobi/license/gurobi.lic")

The command shows some meta-info (name of the module, its version, etc.) When you load this module, it adjusts various environment paths (as you see there), so that when you type python it runs the program from /share/apps/anaconda-ci/fgci-centos7-anaconda/software/anaconda/2023-01/2eea7963/bin. This is almost magic: we can have many versions of any software installed, and everyone can pick what they want, with no conflicts.

Loading modules

Let’s dive right into an example and load a module.

Local differences

If you are not at Aalto, you need to figure out what modules exist for you. The basic princples probably work on almost any cluster.

Let’s assume you’ve written a Python script that is only compatible with Python version 3.7.0 or higher. You open a shell to find out where and what version our Python is. The type program looks up the current detected version of a program - very useful when testing modules (if this doesn’t work, use which).:

$ type python3
python3 is /usr/bin/python3
$ python3 -V
Python 3.6.8

But you need a newer version of Python. To this end, you can load the anaconda module using the module load anaconda command, that has a more up to date Python with lots of libraries already included:

$ module load anaconda
$ type python
python3 is /share/apps/anaconda-ci/fgci-centos7-anaconda/software/anaconda/2023-01/2eea7963/bin/python3
$ python -V
Python 3.10.8

As you see, you now have a newer version of Python, in a different directory.

You can see a list of the all the loaded modules in our working shell using the module list command:

$ module list
Currently Loaded Modules:
  1) anaconda/2023-01

Note

The module load and module list commands can be abbreviated as ml

Let’s use the module purge command to unload all the loaded modules:

$ module purge

Or explicitly unload the anaconda module by using the module unload anaconda command:

$ module unload anaconda

You can load any number of modules in your open shell, your scripts, etc. You could load modules in your ~/.bash_profile, but then it will always automatically load it - this causes unexplainable bugs regularly!

Module versions

What’s the difference between module load anaconda and module load anaconda/2023-01?

The first anaconda loads the version that Lmod assumes to be the latest one - which might change someday! Suddenly, things don’t work anymore and you have to fix them.

The second loading anaconda/2023-01 loads that exact version, which won’t change. Once you want stability (possibly from day one!), it’s usually a good idea to load specific version, so that your environment will stay the same until you are done.

What’s going on under the hood here?

In Linux systems, different environment variables like $PATH and $LD_LIBRARY_PATH help figure out how to run programs. Modules just cleverly manipulate these so that you can find the software you need, even if there are multiple versions available. You can see these variables with the echo command, e.g. echo $PATH.

When you load a module in a shell, the module command changes the current shell’s environment variables, and the environment variables are passed on to all the child processes.

You can explore more with module show NAME.

Making a module collection

There is a basic dependency/conflict system to handle module dependency. Each time you load a module, it resolves all the dependencies. This can result in long loading times or be annoying to do each time you log in to the system. However, there is a solution: module save COLLECTION_NAME and module restore COLLECTION_NAME

Let’s see how to do this in an example.

Let’s say that for compiling / running your program you need:

  • a compiler

  • CMake

  • MPI libraries

  • FFTW libraries

  • BLAS libraries

You could run this each time you want to compile/run your code:

$ module load gcc/9.2.0 cmake/3.15.3 openmpi/3.1.4 fftw/3.3.8-openmpi openblas/0.3.7
$ module list           # 15 modules

Let’s say this environment works for you. Now you can save it with module save MY-ENV-NAME. Then module purge to unload everything. Now, do module restore MY-ENV-NAME:

$ module save my-env
$ module purge
$ module restore my-env
$ module list           # same 15 modules

Generally, it is a good idea to save your modules as a collection to have your desired modules all set up each time you want to re-compile/re-build.

So the subsequent times that you want to compile/build, you simply module restore my-env and this way you can be sure you have the same previous environment.

Note

You may occasionally need to rebuild your collections in case we re-organize things (it will prompt you to rebuild your collection and you simply save it again).

Full reference

Command

Description

module load NAME

load module

module avail

list all modules

module spider NAME

search modules

module list

list currently loaded modules

module show NAME

details on a module

module help NAME

details on a module

module unload NAME

unload a module

module save ALIAS

save module collection to this alias (saved in ~/.lmod.d/)

module savelist

list all saved collections

module describe ALIAS

details on a collection

module restore ALIAS

load saved module collection (faster than loading individually)

module purge

unload all loaded modules (faster than unloading individually)

Final notes

If you have loaded modules when you build/install software, remember to load the same modules when you run the software (also in Slurm jobs). You’ll learn about running jobs later, but the module load should usually be put into the job script.

The modules used to compile and run a program become part of its environment and must always be loaded.

We use the Lmod system and Lmod works by changing environment variables. Thus, they must be sourced by a shell and are only transferred to child processes. Anything that clears the environment clears the loaded modules too. Module loading is done by special functions (not scripts) that are shell-specific and set environment variables.

Triton modules are also available on Aalto Linux: use module load triton-modules to make them available.

Some modules are provided by Aalto Science-IT, and on some clusters they could be provided by others, too. You could even make your own user modules.

Exercises

Before each exercise, run module purge to clear all modules.

If you aren’t at Aalto, many of these things won’t work - you’ll have to check your local documentation for what the equivalents are.

Modules-1: Basics

module avail and check what you see. Find a software that has many different versions available. Load the oldest version.

Modules-2: Modules and PATH

PATH is an environment variable that shows from where programs are run. See it’s current value using echo $PATH.

type is a command line tool (a shell builtin, so your shell may not support it, but bash and zsh do) which tells you the full path of what will be run for a given command name - basically it looks up the command in PATH

  • Run echo $PATH and type python.

  • module load anaconda

  • Re-run echo $PATH and type python. How does it change?

Modules-3: Complex module and PATH

Check the value of $PATH. Then, load the module py-gpaw. List what it loaded. Check the value of PATH again. Why is there so much stuff? Can you find a module command that explains it?

Modules-5: Modules and dependencies

Load a module with many dependencies, such as r-ggplot2 and save it as a collection. Purge your modules, and restore the collection.

See also

What’s next?

The next tutorial covers data storage.