Bug report

When running Nextflow on Google Batch with gcsfuse mounted directories, and attemping to stage in many input files for a task, the task will fail with the following error:

Error executing process > 'COPY_FILES'

Caused by:
  No such file or directory: /mnt/disks/nf-tower-test-eu-1/scratch/1rnFXXfWvpW08n/f7/b2319081b9b7957733eebb846c0fd3/.command.stage

This appears to be related to the fix implemented in #4282 and reported in #4279, which was intended to disable the separate staging script for remote object storage entirely. However, the fix doesn't properly work on Google Batch.

Steps to reproduce the problem

1. Run a Nextflow pipeline on GCP with a task that stages in many input files (e.g., ~1000 or more files

# create random files
for ((n=0;n<6000;n++)); do touch dummy_file_${n}.txt; done

# sync to a GCS bucket
gsutil -m rsync -r ./ gs://nf-tower-test-eu-1/esha/many_files_test/

One process workflow:

process COPY_FILES {
    input:
        path files
    output:
        path("outdir", type: 'dir')
    script:
    """
    mkdir -p outdir
    for f in ${files}; do
        cp \$f outdir/
    done
    """
}

workflow {
    Channel.fromPath(params.input).collect()
    | COPY_FILES()
}

3. Instead of staging in the directory, stage in each individual file which will result in a large .command.run exceeding 1MB.
4. The task fails because it tries to access the .command.stage file which isn't properly created or accessible

Environment

• Nextflow version: 24.10.5
• Seqera Platform Cloud Version 24.3.0-cycle4_803f393

Additional context

(Add any other context about the problem here)