Helixer official docker not finding its own python modules when running in AWS through nextflow

[MatteoSchiavinato]

Describe the bug
I have successfully run Helixer locally using the pre-constructed docker container. Note that I am running it without GPU. When I run it on AWS using Nextflow, this happens:

  Traceback (most recent call last):
    File "/home/helixer_user/Helixer/scripts/fetch_helixer_models.py", line 5, in <module>
      from helixer.core.data import prioritized_models, fetch_and_organize_models
  ModuleNotFoundError: No module named 'helixer'

To Reproduce
This is my nextflow process:

process HELIXER {

    input:
    tuple \
    val(alias),
    path(fasta)

    val species
    val lineage

    output:
    tuple \
    val(alias),
    path(fasta),
    path("${alias}.gff3"), emit: models

    script:
    """
    set -e 
    set -o pipefail 

    # change directory so that modules can be found properly 
    cd /home/helixer_user/Helixer

    # download models 
    /home/helixer_user/Helixer/scripts/fetch_helixer_models.py

    # add helixer path to pythonpath 
    export PYTHONPATH="/home/helixer_user/Helixer:\${PYTHONPATH}"

    # run helixer 
    /home/helixer_user/Helixer/Helixer.py \
    --compression lzf \
    --fasta-path ${fasta} \
    --gff-output-path ${alias}.gff3 \
    --lineage ${lineage} \
    --min-coding-length 60 \
    --species ${species}
    """

To reproduce the behavior, all that is needed is to run this process using the preconstructed container as container for the process, with no modification.

Error
The error I'm getting:

  Traceback (most recent call last):
    File "/home/helixer_user/Helixer/scripts/fetch_helixer_models.py", line 5, in <module>
      from helixer.core.data import prioritized_models, fetch_and_organize_models
  ModuleNotFoundError: No module named 'helixer'

Seems to come from the fact that the environment does not find the helixer directory from where to load the local modules prioritized_models and fetch_and_organize_models. Even if I change directory into the one where this folder is, or if I add it to the PYTHONPATH (see the nextflow process above).

Note that this behavior is not observed when running locally, just in AWS. AWS + Nextflow notoriously creates some issues when something is installed in /usr, i.e. chances are this something won't be found when running in the cloud (not always, but frequently).

Environment (please complete the following information):
The official docker from helixer

Suggested fix
I am adding a suggestion to fix it. I haven't fully tested it, but it should work if everything (including CUDA drivers) is installed outside of /usr, preferrably in /opt.

Here's the changes I've done to the docker file:

FROM gglyptodon/helixer-docker:helixer_v0.3.3_cuda_11.8.0-cudnn8
USER root 
WORKDIR /opt
RUN mkdir -p /opt/Helixer_inst \
    && mv /home/helixer_user/* /opt/Helixer_inst/
WORKDIR /opt/Helixer_inst/Helixer
ENV PYTHONPATH="/opt/Helixer_inst/Helixer:${PYTHONPATH}"
CMD ["bash"]

This goes for the python packages too. In fact, it would be important to install them outside of /usr/local, e.g. just /usr.

I have solved at least that issue by setting this in the dockerfile:

ENV PYTHONUSERBASE=/usr 

[felicitas215]

Hi,
thank you for reporting this issue. Can you try setting the Python path before evoking the first python script?
Like this:

# add helixer path to pythonpath 
export PYTHONPATH="/home/helixer_user/Helixer:\${PYTHONPATH}"

# download models 
/home/helixer_user/Helixer/scripts/fetch_helixer_models.py

[MatteoSchiavinato]

That was my first attempt, but wasn't successful.

[MatteoSchiavinato]

Here's a dockerfile I've used to construct an image that fits all the criteria above. I'm attaching it so that you may see if anything is useful for you! It creates a very heavy container (21.9 GB), it would need lots of improvement and perhaps a multi-stage build to reduce size (your original is 7.74 GB). But anyway, thought it was useful.

Dockerfile.txt

[MatteoSchiavinato]

UPDATE: I've managed to reduce the docker to about 12 GB based on your new commit and some other changes. Here it is (attached).

Dockerfile_v2.txt

[felicitas215]

Could you try adding RUN pip install --upgrade pip to your Dockerfile. There seems to be an issue with Ubuntu 22.04 not recognizing name and version and installing UNKNOWN 0.0.0 instead of for example helixer-0.3.4. (Link to the bug report: pypa/setuptools#3269).

[felicitas215]

I'm closing this issue, because there was no activity for 1 month. Feel free to reopen the issue if necessary.