Problem:

How to set up Conda (or Miniconda) for shared environments between users and without duplicating heavy packages across their home directories?

Solution:

Install Conda in a centralized directory, configure a shared environment path via a .condarc file, and manage Linux/Windows user group permissions.

Step 1: Install Conda Globally

An administrator must install Conda (or Miniconda) to a public directory accessible to all users. Avoid installing it in any single user’s /home directory. Conda Documentation.

# Download and install to a shared directory like /opt/conda
sudo bash Miniconda3-latest-Linux-x86_64.sh -b -p /opt/conda

Use code with caution.

Step 2: Configure System-Wide Paths

Create or modify the global configuration file at /opt/conda/.condarc. This forces Conda to look for and create environments in your designated shared location. Conda Documentation.

# /opt/conda/.condarc
envs_dirs:
  - /opt/conda/envs
  - ~/.conda/envs

pkgs_dirs:
  - /opt/conda/pkgs
  - ~/.conda/pkgs

The first paths listed act as the primary shared target. The second paths serve as user-specific fallbacks. Conda Documentation.

Step 3: Set Group Directory Permissions

To let team members utilize or update these environments, manage permissions using a dedicated system group. Anaconda.

  1. Create a user group:

    sudo groupadd bioinformatics

  2. Add users to the group:

    sudo usermod -aG bioinformatics username

  3. Change directory ownership:

    sudo chown -R root:bioinformatics /opt/conda

  4. Set Read/Write permissions:
    Choose one of the two strategies below depending on how much control you want to grant your users:

    • Admin-Managed (Read-Only for users): If only administrators should modify environments, leave permissions restrictive. Users can still run conda activate but cannot change packages.

    • Collaborative (Read/Write for users): If users need to modify environments or install packages, run:

      sudo chmod -R 775 /opt/conda
sudo chmod -R g+s /opt/conda/envs /opt/conda/pkgs

      (The g+s SGID bit ensures new folders inherit the bioinformatics group ownership automatically). GitHub

Step 4: Configure centralized initialization script

sudo ln -s /opt/conda/etc/profile.d/conda.sh /etc/profile.d/conda.sh

Explanation:

It recommends creating a symbolic link (ln -s) to automate the setup for all current and future users simultaneously, rather than making every individual user manually edit their personal ~/.bashrc file.

Here is why this method is highly efficient for system administrators:

1. Global Automation

The /etc/profile.d/ directory is a special folder used by Linux operating systems. Every time any user logs into the system or opens a new terminal shell, Linux automatically runs every script located inside this folder. By placing the Conda initialization shortcut there, Conda becomes instantly available to everyone.

2. No Manual User Intervention

Without this link, every new team member or user added to the system has to manually run source /opt/conda/etc/profile.d/conda.sh or conda init. Linking it globally removes human error and ensures the system works right out of the box for everyone.

3. Clean and Maintainable

A symbolic link is just a shortcut. If you ever update Conda or change files inside /opt/conda/, you do not have to update any user files. The shortcut simply continues to point to the correct, active configuration.

(option 2) Configure User Profiles

Every user needs access to the centralized Conda initialization script. Instruct your users to add this command to their ~/.bashrc or profile initialization files: Stack Overflow

source /opt/conda/etc/profile.d/conda.sh

Once updated, running conda env list will show the shared environments to all members of the group.

Comparison of the Two Approaches

  • The Global Method (The Symlink): The administrator runs the sudo ln -s command once. The system handles the rest for all users automatically.

  • The User Method (Step 4): Every user must manually add source /opt/conda/etc/profile.d/conda.sh to their own ~/.bashrc file.

Sources: