Removal of packages from anaconda "defaults" channel in user environments on JASMIN

Background  

Following a change in the licensing conditions by Anaconda, it is now the case that all users of the Anaconda defaults Conda package channel (repository) from organisations of 500 or more employees are potentially liable to pay for usage, even if it is for the purpose of academic research.

This does not affect packages from the community channel conda-forge, which remains free to use, as also does the conda installer program itself.

It has to be assumed that JASMIN users in general would potentially be subject to contractual liability if Conda packages from the defaults channel are used, so it has been decided that these are not to be used on JASMIN.

The JASMIN team have now taken steps to ensure that centrally-provided environments including Jaspy make use only of conda-forge, but as regards packages in users’ own directories, the responsibility falls on individual users to do the same.

These days, the miniforge installer is available to install conda environments, and this will install packages from conda-forge by default, but we have found there to be many user conda environments on JASMIN which contain packages from defaults – either because these environments pre-date the use of miniforge, or because the channel was specified explicitly during installation – and these need to be addressed.

Because of the sometimes complex dependencies between packages in conda channels, it is difficult to automate the removal of packages from the defaults channel, if the desired end result is a usable conda environment containing equivalent packages from conda-forge. Some manual decision-making may be needed, and this document is a guide to help you to do this.

Please note that this page is a best-efforts guide only. Ultimately you are responsible for the contents of the conda environments in your own user directories (including group workspaces, etc). We cannot warrant that following these instructions will succeed in removing all the packages for which you could incur charges, despite that being the intention.

What conda environments do I have?  

You can generally discover which conda environments you have created, by typing conda env list (when you have an environment activated), or by looking in your ~/.conda/environments.txt file. Occasionally, for some reason some environments might not be listed, so here are some other likely places where you might find conda environments that you have created:

• Under ~/miniconda3/envs or ~/anaconda3/envs or ~/miniconda3/envs or ~/mambaforge3/envs.
• Also variants of the above without the 3, or with a 2 in place of the 3 (for environments created using the Python 2 installer versions).
• Under ~/.conda/envs (note the dot).
• (If you have used a custom directory), in other subdirectories of the same envs directories as used by the other conda environments that are displayed by conda env list.

Note that if you have created named conda environments, you will usually also have the associated installer base environments. For example, if you have environment ~/miniconda3/envs/myenv then the base environment is at ~/miniconda3.

Which channels do my installed conda packages use?  

There are various ways to list the contents of an environment: the names of the packages and the channels they come from. Select one of the following:

• To get a list of packages with their channel names, activate the environment and then type conda list. However, note that if the channel is not mentioned in the output, then the package is from the defaults channel (also known as main). The list will also include any pip-installed packages, and these will say pypi instead of a conda channel.

• Or to display the channel URL for each of the packages, inspect the JSON files in the conda-meta subdirectory of the environment. After activating the environment (which sets the CONDA_PREFIX variable), you could type:

grep '"channel"' $CONDA_PREFIX/conda-meta/*.json

• There is also a file called $CONDA_PREFIX/conda-meta/history containing: the commands that you used in order to install packages (see lines beginning with # cmd), which packages were installed (lines beginning +), and which were removed (lines beginning -). In each case, the channel name is shown along with the package.

Whichever of these you do, any packages that are from the main / defaults channel are part of the paid offering, and will need to be removed. If any exist, these should show up in the list of channel URLs as being under https://repo.anaconda.com/pkgs/main/ (or https://repo.anaconda.com/pkgs/r/ for the R language packages).

However, especially because the conda list output does not list a channel name where it is defaults, the safest approach may be to list everything that is not from conda-forge, and then remove these packages unless they are from known free channels. So after activating the environment, you could type:

conda list | egrep -vw `conda-forge|pypi`

(Note that if the environment is already clear of any such packages, you will still see the header lines in the output of this command, so if you see no output at all then something went wrong.)

How do I replace packages that use the defaults channel?  

Once you have identified which environments you have and which packages in them are from the main / defaults channel, here is how to go about replacing these packages with equivalents from conda-forge.

Before starting on this procedure, check whether you have a file called ~/.condarc. If you do, and if it contains a line that references the defaults channel, then remove that line.

Base environments  

First, let’s deal with the base environments.

Your approach to the base environments will depend on whether these were installed using the miniforge/mambaforge installers, or miniconda/anaconda.

miniforge / mambaforge base environments  

You should hopefully have found that it is already true that any such base environments only contain packages from the conda-forge channel, so that you do not have to do anything with these. (In the unlikely event that this is not the case, then deal with them as per the advice for named environments given below.) Furthermore, the conda or mamba commands that they provide in order to install other environments should default to only using conda-forge (provided that you don’t have a ~/.condarc file that overrides this).

miniconda / anaconda base environments  

(You can skip this subsection if you don’t have any miniconda/anaconda base environments.)

By contrast to the above, these base environments will contain packages from the defaults channel, and also, the conda command in these environments will by default try to install packages from that channel. For this reason, we recommend that these base environments should be deleted completely.

If the base environment does not contain (inside its envs subdirectory) any named environments that you wish to keep, then you can simply delete it, for example: rm -fr ~/miniconda3

However, if it does, then you can keep the envs subdirectory and delete the subdirectories other than envs, for example by doing:

cd ~/miniconda3
mkdir to-delete
mv * to-delete/
mv to-delete/envs ./
ls to-delete  # check what we are about to delete
rm -rf to-delete

You won’t need the miniconda base environment just in order to use one of the named sub-environments inside it, because you can instead use a miniforge base environment to provide the conda and mamba executables. If you do not already have miniforge installed, you can get the installer here  .

Assuming that you have installed a miniforge base environment at ~/miniforge3, you can activate one of your old miniconda environments by specifying its full path after activating the base environment. For example, this might look like:

source ~/miniforge3/bin/activate
conda activate ~/miniconda3/envs/myenv

(Do not be tempted just to move the environment to a new path under ~/miniforge3/envs in order to avoid the need to type the full path, because it will probably contain hard-coded paths which would not work if it is moved. However, you could create a symbolic link.)

Remember that in the sub-environments you are keeping, you will need to purge any packages that use the paid channels, as described below.

Named environments  

By this point, your only base environment(s) should be miniforge (or mambaforge) environments, but you may have a number of named environments that contain packages from the defaults channel, which need dealing with.

Mostly, any such named environments will be ones that were created using miniconda (which you will now activate via the miniforge base environment as described above). However it could still arise that you have mambaforge environments containing packages from defaults, if this was specified either in your .condarc file or in a .yml file that you used to create the environment.

How you deal with these environments will probably depend on the number of packages from the defaults channel that they contain. We would suggest that if they only contain a small number, then you can attempt to patch the existing environment, but if they contain a larger number then it will be best to create a new environment with similar contents but from conda-forge. Here are the details:

Patching an existing environment  

This is a suggested approach where you have an environment that only has a small number of packages that are from the defaults channel.

• Activate the environment.

• Try typing conda remove followed on the command line by a list of the affected packages, and see which packages it proposes to remove, bearing in mind that if other packages depend on the ones in question, they will be included also.

• If the list is reasonably short, then confirm the changes, and after that, reinstall the packages, including any dependent packages, using

  conda install -c conda-forge --override-channels package1 [package2...]

  substituting here the names of the relevant packages. (The installation options shown above should make sure that new the packages come from conda-forge, even if there is still some .condarc file that says otherwise.)

  You can also use mamba instead of conda here, to use the mamba installer.

  Once you have done this, remember to recheck the list of packages and channels. (For example, if you do conda list, do they now all show up as being from conda-forge?)

• If the list of dependencies is unacceptably long, then answer no. You will probably have to recreate the environment instead, as shown below. (You could perhaps instead attempt to reinstall the relevant packages by adding the --force-reinstall option on the conda install command instead of a separate conda remove step first, but when we tried an example, this did not succeed – there may be dependencies on exact release versions, for example.)

Recreating an environment  

This suggestion is for where too many changes are needed to be able to modify the existing environment easily.

It is important to note that although there is a possible procedure for obtaining an exact copy of an environment (namely: export a list of packages to a file including all the exact version and release numbers, and then use it to create the new environment), the aim here is, rather, to produce an equivalent environment based on packages from conda-forge. Package releases will differ slightly from what is available in the defaults channel. So it will be best to be avoid constraining the requirements too rigidly, in order give the installer the flexibility it needs to choose mutually compatible versions of all of the packages.

The aim, therefore, will be to repeat the steps that you used (for example, install matplotlib), rather than try to replicate the exact versions that you ended up with.

You can see what conda install commands you ran previously, by going to the environment directory of the old environment (if you have activated the environment, this will be at $CONDA_PREFIX), and typing:

grep "# cmd" conda-meta/history

In addition to any conda install commands, it is possible that you also installed Python packages using pip install. The specific commands will probably not have been recorded (other than the limited record in your ~/.bash_history file), but if you activate the old environment and type pip freeze, this shows which Python packages are installed. To see just the ones that were not installed via conda packages, type:

pip freeze | grep -v " @ "

Note that this list will contain any packages that were installed automatically for dependencies, in addition to the ones that you installed explicitly.

Once you have obtained this information, you are in a position to repeat the installation. First ensure that you have deactivated any other conda environments in your session, and then activate your miniforge base environment. Depending where this is located, a typical example would be:

source activate ~/miniforge3/bin/activate

Then you will repeat the conda install commands that you typed – or optionally, replace conda create with mamba create to use the (more efficient) mamba installer (and likewise with mamba env create and mamba create where appropriate). Note the following:

• If you created the old environment using conda env create and a .yml file, and you still have this file available, then before reusing it, look at the channels section (usually near the top), and edit to to ensure that conda-forge is listed and defaults is not.

  If you used conda env create but no longer have the .yml file, then see below for an alternative approach.

• After you have created the environment, remember to activate it before issuing any later conda install (or mamba install) commands. Otherwise you will end up installing the packages into your base environment instead.

Then install the additional pip packages, using pip install [package_name]. Again, this should be after activating the environment. Note that:

• You only need to name the packages that your code is importing explicitly; any dependencies will be installed for you.

• Remove the version requirements (e.g. pip install dask instead of pip install dask=2023.2.0) if you wish to install a recent version instead of the originally used version.

• If you created the environment from a .yml file that contains a pip section, then some or all of these may already have been installed for you by this stage.

• You can also put multiple package requirements into a file (typically called requirements.txt) and install them using pip install -r [filename]. For example, it might be more convenient to do this if you choose to reinstall everything that was reported when you ran the pip freeze | grep -v " @ " command in the old environment, rather than a selected few packages.

Once you have got your new environment installed and working, remember to delete the old one.

What to do if a previously used YAML file is no longer available  

If you do not have a record of packages based on the conda history (for example, because they refer to a .yml file that no longer exists), then you will need to start from the list of installed packages in the old conda environment.

To do this, start by activating the old environment, and then export the package list to a file:

conda env export > environment.yml

You will use this file to create the new environment. But first, you should edit the file:

• In the channels section, ensure that conda-forge is included and defaults is not.

• Then go through and simplify it as much as possible, so that it contains the list of necessary packages but does not constrain the environment too tightly, for reasons discussed above:

  • Remove packages that you do not recognise; only include what you will use explicitly. Remember that any dependencies will get added for you automatically. This applies both to the list of conda packages and also any packages from PyPI if the file contains a pip: section.
  • Relax the version numbers: definitely remove the exact release (after the second = sign), and also consider removing the package version number – or possibly changing it to >= to permit also more recent versions. For example, change sqlite=3.41.2=h5eee18b_0 to just sqlite or sqlite>=3.41.2.

Once you have done this, activate your miniforge base environment, and then use it create your new environment, for example:

conda env create -n my_new_env -f environment.yml

or using the mamba installer:

mamba env create -n my_new_env -f environment.yml

As above, remember to delete your old environment afterwards.