Install a software package

In Nuvolos, almost all applications are equipped with conda environments and we also let our users load user-written packages and extensions.

Basic provisions

All Nuvolos applications come with a set of useful *nix applications pre-installed:

• git

• git-lfs

• vim

• build-essential (includes make)

• xterm for GUI based applications for terminal emulation

• Nuvolos data connectors for supported languages

The conda environment

Most Nuvolos applications come equipped with the package manager conda, and more recent applications alias conda to a faster drop-in replacement called mamba. Conda is a language-agnostic package manager which lets you install language-specific packages and system libraries as a non-root user. For the Python language, most packages available via pip can also be installed via conda.

Try to install software with conda first and keep pip as a fallback option. Exception to this rule are GPU related libraries, please consult our GPU documentation on the topic.

We also recommend passing the '--freeze-installed' flag when installing with conda, to ensure the minimal possible changes to the conda environment.

If you cannot self-service your packages, contact us at [e-mail] and we will help you.

As an example, suppose you want to install imagemagick and gifsicle for mass editing gifs. The following command will install this to the conda environment of your application:

conda install --freeze-installed gifsicle imagemagick

When distributing and snapshotting an application, the contents of the conda environment are also impacted. This is a key feature for reproducibility.

LaTeX

The most widespread installers of TeXLive require root privileges to install TeX, and thus won't work with Nuvolos applications. You have a handful of options to work with TeX on Nuvolos:

1. You can install Overleaf as a standalone Nuvolos application

2. You can install various JupyterLab versions with TeX preinstalled (search by the tag language:latex)

3. You can install TeX using TinyTex

Install TinyTeX

Open a terminal in your application, and execute the following command:

export TINYTEX_DIR=${CONDA_DIR:-~} && wget -qO- "https://yihui.org/tinytex/install-bin-unix.sh" | sh && if [ -f /files/.nuvolos/.bashrc ]; then echo "PATH=$TINYTEX_DIR/.TinyTeX/bin/x86_64-linux:$PATH" >> /files/.bashrc; else echo "PATH=$TINYTEX_DIR/.TinyTeX/bin/x86_64-linux:$PATH" >> /files/.nuvolos/.bashrc; fi && export PATH=$TINYTEX_DIR/.TinyTeX/bin/x86_64-linux:$PATH && echo "Successfully installed TinyTex"

If your application has a conda environment, TinyTeX will be installed next to it, and that means your TeX environment will be snapshotted/distributed along with your regular conda packages

TinyTex installs by default a lightweight TeX distribution. You can use tlmgr to install additional TeX packages you need, e.g.

tlmgr install collection-fontsrecommended

You can list available collections with

tlmgr info --list collections

and schemes with

tlmgr info --list schemes

TinyTeX in notebooks

To use TeX from notebooks, you'll probably need to execute the following command in one of the first cells of your notebook

import os
os.environ['PATH'] = f"{os.environ.get('CONDA_DIR') or os.path.expanduser('~')}/.TinyTeX/bin/x86_64-linux:{os.environ['PATH']}"

Tips and tricks

Single-purpose applications

We strongly suggest creating single-purpose applications.

This practice has the following benefits:

1. Conda or R package environments remain monolithic and fairly lightweight. Adding more and more packages to the same environment will inevitable result in an unmanageable environment.

2. Distribution and snapshotting takes less storage and resources and conclude faster.

Create a new application instead of upgrading

If you are contemplating doing a major version update on your application, we suggest creating a new app in the same instance and starting there from scratch.

This practice has the following benefits:

1. Conda environments can break after major updates.

2. The reproducibility of your work may suffer - however it is trivial to maintain two monolithic and separate application structures in parallel, even in the same instance!

3. Distribution is based on filesystem-differences and after-upgrade distributions may become less stable due to the massive number of changes occurring on the filesystem.