Hi ACCESS-NRI hard working people,
The new payu version is interfering with my creation of conda environments, for which I use mamba as a faster alternative to conda.
When I enter into Gadi, I always do the following:
module use /g/data/vk83/modules
module load payu
If I run:
mamba env create -f my_env.yml
I get the following error:
bash: /g/data/vk83/apps/base_conda/bin/mamba: No such file or directory
Do you know what’s happening?
Pearse
FWIW I generally do
module unload payu
whenever I’m using conda, because of similar interference with conda analysis modules.
Yeah, the Payu module uses a shared micromamba installation in /g/data/vk83. So you’ll need to unload the module as @dkhutch suggests.
As a general rule, it’s a good idea to only load modules when you need them to avoid these sorts of conflicts.
Would it be more user-friendly for the payu module to only expose the payu command rather than including all the conda stuff? E.g. by putting a symlink to payu in a new directory and having the module only add that directory to $PATH.
^ ping @jo-basevi
The latest payu module is deployed using Dale Robert’s work at CLEX of containerising conda environments (forked from the cms-conda-singularity repository) so it will be similar to the analysis environments. The module sets environment variables to what it would be if micromamba activate was run on the un-containerised version of the conda environment. An exception is the$PATH outside the container includes the directory of launcher entry-point scripts (e.g. payu) that launch the container and runs the command inside the container. It might be possible to remove the conda-specific environment variables, though I’d have to test to see if anything breaks (created related issue here).