Bug report

Expected behavior and actual behavior

When using the conda directive for a process to point to a environment yaml file, the environment gets re-created in the cache directory on every run when using Seqera Platform.
We would expect that nextflow would use the existing conda environment in the cache directory.

This change in behaviour is due to the conda cache changes here: #5489

Steps to reproduce the problem

1. create a nextflow pipeline with the following process and environment file

process foo {
  conda '/some/path/my-env.yaml'

  script:
  """
  your_command --here
  """
}

name: my-env
channels:
  - conda-forge
  - bioconda
dependencies:
  - star=2.5.4a
  - bwa=0.7.15

2. Import this pipeline into Seqera Platform - https://docs.seqera.io/platform-cloud/getting-started/quickstart-demo/add-pipelines
3. Configure the `nextflow.config' of pipeline to:
   • enable conda conda.enabled = true
   • set the conda.cacheDir = /some/path/on/the/system
4. Launch the pipeline via Seqera (in my case I am using a SLURM compute environment)
   • Seqera will:
     • set NXF_ASSETS to a location in the work directory <basepath>/work/.nextflow/pipelines/<pipeline-id>
     • launch nextflow on the server with nextflow run <pipeline> ... -with-tower <seqera-url>

Program output

Monitor the execution with Seqera Platform using this URL: <seqera-url>
Creating env using conda: /home/user/<username>/nextflow/work/.nextflow/pipelines/f1f16f92/my-pipeline/envs/environment.yml [cache /home/user/<username>/nextflow/conda_cache/env-fe50773488a5e0e5-535eefbee5734a58afed514fbe57635c]

Environment

• Nextflow version: 24.10.5
• Java version: 17.0.14
• Operating system: Linux
• Bash version: 5.1.8

Additional context

My troubleshooting has lead me to understand that Seqera Platform generates a new directory for nextflow assets for every pipeline launch. The new directory is placed in the work/.nextflow/pipelines directory. This directory has a new name for every pipeline launch.
In my case:

• /home/user/<username>/nextflow/work/.nextflow/pipelines/f1f16f92/ - Run 1
• /home/user/<username>/nextflow/work/.nextflow/pipelines/ee5ee5c/ - Run 2

This is problematic because the CondaCache class computes the cached environment directory based on a hash of the full path of the environment.yml file:

• /home/user/<username>/nextflow/work/.nextflow/pipelines/f1f16f92/my-pipeline/envs/environment.yml becomes env-fe50773488a5e0e5-535eefbee5734a58afed514fbe57635c

Essentially there is no way for two pipeline runs in Seqera to use a cached conda environment because they will never have the same name.
I see this as an incompatibility between the way Seqera Platform uses nextflow, so I'm not 100% sure where the fix should lie.

A current workaround may be to specify conda requirements within a string (instead of file):

process foo {
  conda 'star=2.5.4a bwa=0.7.15'
  ...
}

An alternative way (within nextflow) may be to make the generation of the conda environment hash more configurable: (ie. only take path from project root).