EcoliTyper tries to run analyses from the conda environment

[mariemoinet]

Hi,
I gave a go at this program on a handful of sequences, but can't get it to work.
We have conda environments running inside containers that are by design read only.
I get the same error for all modules (eg below for serotyping and test.fna):
[91m[1m❌ ERROR:[0m
[91m Serotyping analysis failed: [Errno 30] Read-only file system:[0m
[91m '/opt/conda/envs/ecolityper/lib/python3.12/site-[0m
[91m packages/ecoliTyper/modules/serotypefinder_module/test.fna'[0m

I'm not sure why it would be wanting to write to a python lib?
Is there a way to get it to write on the working dir or another type of tmp dir instead?

Also, it seems the stderr is redirected to stdout, my stderr log file was empty while the stdout contained all errors. This is not very helpful imho...

[mariemoinet]

as an add-on: our HPC engineer (who has write rights on conda environments) had a try himself: "even if you let it write to the modules dir, it doesn't clean up like it's supposed to so the next person will run their analysis on your files as well as their own as it just processes every file in the dir".
This is definitely something that needs correcting.

[russellsmithies]

If you use the builtin python tempfile functions then it takes care of cleaning up automatically. I'd suggest having a a $TMPDIR environment variable to set the location for the temp files because not everyone is using Linux.

[bbeckley-hub]

If you use the builtin python tempfile functions then it takes care of cleaning up automatically. I'd suggest having a a $TMPDIR environment variable to set the location for the temp files because not everyone is using Linux.

Thanks for this @russellsmithies

[bbeckley-hub]

@mariemoinet Thank you again for the detailed report and for helping me debug this.

Confirmed root cause

The orchestrator copies every input FASTA file into each module’s own installation directory (e.g., .../site-packages/ecoliTyper/modules/serotypefinder_module/). It then runs the module with that directory as the current working directory.

• In a normal Linux Conda environment, site-packages is writable, so this works.
• In your read‑only container, that directory is read‑only, so the copy fails with [Errno 30].

Some modules also write temporary files (like test.fna) into the same directory, which adds to the problem.

Immediate workarounds (choose one)

1. Bind‑mount a writable directory over the modules folder (no code changes)

# On the host, create a writable directory
mkdir -p /tmp/ecolityper_writable/modules

# Mount it into the container, hiding the read‑only original
docker run -v /tmp/ecolityper_writable/modules:/opt/conda/envs/ecolityper/lib/python3.12/site-packages/ecoliTyper/modules ...

Important: You must manually clean /tmp/ecolityper_writable between runs, or leftover files will be reprocessed.

2. Run the tool outside the container (simplest & most reliable)

Run EcoliTyper on a regular Linux machine or your HPC’s login node (where the Conda environment is writable), then copy the results back.

3. Make the Conda environment writable (if your HPC allows)

conda create -p /some/writable/path/ecolityper -c conda-forge -c bioconda -c bbeckley-hub ecolityper
conda activate /some/writable/path/ecolityper

Permanent fix (next release)

I am refactoring the whole pipeline to:

1. Create a single global temporary directory per run using tempfile.mkdtemp() as suggested by @russellsmithies . This directory will be writable and automatically cleaned up.
2. Stop copying files into module directories – instead, copy inputs to the global temp directory.
3. Modify all modules to accept a --tmp-dir argument (or reuse the existing --output) and write all files (including temporary ones) to that directory.
4. Remove the orchestrator’s cleanup_module_directory (no longer needed).

After this fix, the tool will work correctly in read‑only Conda environments, containers, and multi‑user systems without cross‑contamination.

I’ll push the updated version within a week. Thank you for your patience and for helping improve EcoliTyper.

@russellsmithies Thanks again for the tempfile suggestion – I’ll implement it.