env config not working with GCP Batch

[nick-youngblut]

Bug report

My nextflow.config contains:

env {
    MY_ENV_VAR = "my_value"
}

When the executor is local, the process using MY_ENV_VAR via os.environ["MY_ENV_VAR"] runs successfully.
However, when the executor is google-batch, the process throws KeyError: 'MY_ENV_VAR'.

https://www.nextflow.io/docs/latest/reference/config.html#env does not state that env is not supported for google-batch, so I'm assuming that there is either a bug or a lack of docs.

Expected behavior and actual behavior

Environmental variables set via the env config scope should be available for GCP Batch jobs... or the docs should explicitly state that env is not supported for GCP Batch.

Steps to reproduce the problem

• Set variables in the env scope
• Run locally; processes using the env variable should succeed
• Run on GCP Batch; processes using the env variable should fail

Environment

• Nextflow version: 24.10.2
• Java version: 21.0.0
• Operating system: Ubuntu 22.04.5
• Bash version: 5.1.16

[pditommaso]

I'm unable to replicate this. Can you please provide a test case ?

[nick-youngblut]

My reproducible example:

main.nf

workflow { 
    PRINT_ENV()   
    PRINT_ENV_PY()  
}

process PRINT_ENV {
    publishDir file(params.output_dir),  mode: "copy", overwrite: true
    container "ubuntu:latest"
    
    output:
    path "out.txt"

    script:
    """
    echo "$my_env_var" > out.txt
    """
}

process PRINT_ENV_PY {
    publishDir file(params.output_dir),  mode: "copy", overwrite: true
    container "python:3.11"
    
    output:
    path "out_py.txt"

    script:
    """
    test.py > out_py.txt
    """
}

nextflow.config

params {
    output_dir = "pipeline_output"
}

env {
    my_env_var = "TEST"
}

profiles {
    gcp {
        workDir            = "gs://"###/sandbox/work"
        fusion.enabled     = false
        wave.enabled       = false
        params.output_dir  = "gs://"###/sandbox/pipeline_output"
        process {
            executor       = "google-batch"
            errorStrategy  = "retry"
            maxRetries     = 2
            scratch        = true
        }
        google {
            project   = ""###"
            location  = ""###"
            batch {
                serviceAccountEmail = "###"
                spot                             = true
                maxSpotAttempts     = 3
            }
        }
    }
}

test.py

#!/usr/bin/env python3
import os

if __name__ == '__main__':
    print(os.getenv("my_env_var"))

Results

out.txt will always contain TEST when run locally and on GCP Batch. However, out_py.txt will contain TEST when run locally and None when run on GCP.

Notes

If the following is used:

process PRINT_ENV_PY {
    publishDir file(params.output_dir),  mode: "copy", overwrite: true
    container "python:3.11"
    
    output:
    path "out_py.txt"

    script:
    """
    export my_env_var="$my_env_var"
    test.py > out_py.txt
    """
}

Then, the env variable value ("TEST") is written out to out_py.txt. Importantly, exporting the env variable is only needed when running the pipeline on GCP Batch (versus local).