"Permission denied" when running opensafely locally

[inglesp]

I've just tried opensafely run locally for the first time, and I'm getting a Permission denied error.

$ opensafely run produce_breakdown

Running actions: produce_breakdown

jobrunner.run loop started
produce_breakdown: Preparing
produce_breakdown: Copying in code from /home/inglesp/work/ebmdatalab/opensafely/research/imd-breakdown
produce_breakdown: Copying input file generate_study_population: output/input.csv
produce_breakdown: Started
produce_breakdown: View live logs using: docker logs -f job-imd-breakdown-produce-breakdown-chpoalvro7oqt7f7
produce_breakdown: Running
produce_breakdown: Finished, checking status and extracting outputs
produce_breakdown: Logs written to: /home/inglesp/work/ebmdatalab/opensafely/research/imd-breakdown/metadata/produce_breakdown.log
produce_breakdown: Job exited with an error code
produce_breakdown: Cleaning up container and volume

=> produce_breakdown
   Job exited with an error code

   log file: metadata/produce_breakdown.log
   outputs:
     (no outputs)

   logs:

     [FATAL tini (7)] exec analysis/breakdown.py failed: Permission denied

Here's my project.yaml:

version: '3.0'

expectations:
  population_size: 1000

actions:

  generate_study_population:
    run: cohortextractor:latest generate_cohort --study-definition study_definition
    outputs:
      highly_sensitive:
        cohort: output/input.csv

  produce_breakdown:
    run: python:latest analysis/breakdown.py
    needs: [generate_study_population]
    outputs:
      moderately_sensitive:
        log: output/breakdown.csv

It's not clear what I'm doing wrong. Any ideas?

[evansd]

The issue here is that you need:

run: python:latest python analysis/breakdown.py

because analysis/breakdown.py isn't executable and there's no entrypoint script in the associated docker image.

There's an open question about exactly what we want to do here (I thought there was an issue somewhere but I can't now find it). We could give every docker image an entrypoint script which does the "obvious right thing" as we do for e.g. Stata. In this case that would be to run the script as Python. However the trouble is that we sometimes want to do something other than that e.g. invoke Python with special arguments, run Python under a profiler, use jupyter as the executable.

I don't think we've quite got to the bottom of how we do this yet.

[alexwalkerepi]

When I've run python scripts with the job runner before, the run command I've used would be python:latest python analysis/breakdown.py. I'm not sure if the extra python makes a difference here?