如何访问snakemake python 代码中的重试尝试？

Question

When you execute a snakemake script with --restart-times >= 1 it will try to re-execute a failed run. Upon re-execution it is possible to access the number of execution attempts via a lambda function in "resources". However, I would like to access the number of attempts in a block of python code outside of my rule. I have tried to pass the attempt variable from the resources block to my python function, but to no avail. My snakemake version is 5.32.1 and a quick test with 6.0.3 looks very similar.

def getTargetFiles(files, attempted):
    do stuff
    return modified-target-files

rule do_things_rule:
    input: 
        ...
    output:
        getTargetFiles("file/path.txt", resources.attempt)
    resources:
        attempt=lambda wildcards, attempt: attempt,

This unfortunately yields an error. "NameError in line 172 of xxxx.py: name 'resources' is not defined"

The closest I have come, is to access "workflow.attempt" but this seems to be always set to 1. Perhaps this is the default value for attempts?

rule do_things_rule:
    input: 
        ...
    output:
        getTargetFiles("file/path.txt", workflow.attempt)

I was taking a look at the internals of snakemake in the hope of finding a solution. Unfortunately my python knowledge isn't up to the task. There are some variables one can access in place of workflow.attempt, which do not have integer values. Not sure if there is a way of getting the current number of attempts using these slightly differently:

print snakemake.jobs.Job.attempt
<property object at 0x7f4eecba66d0>

print snakemake.jobs.Job._attempt
<member '_attempt' of 'Job' objects>

Answer 1

Here is a minimal working example, with which I could reproduce your error.

def getTargetFiles(files, attempted):
  return f"{files[:-4]}-{attempted}.txt"

rule do_things_rule:
  resources:
    nr = lambda wildcards, attempt: attempt
  output:
    getTargetFiles("test.txt", resources.nr)
  shell:
    'echo "Failing on purpose to produce file'
    '{output} at attempt {resources.nr}'
    '"; exit 1 '

Indeed, output does not know resources . I assume that this is because it needs to be accessed before the rule has run (see below). In contrast, if you replace getTargetFiles("test.txt", resources.nr) by getTargetFiles("test.txt", 1) , then the rule runs the right number of times and the shell command has access to resources.nr .

As far as I understand, there is a fundamental reason for this problem.

A snakemake workflow is "defined in terms of rules that define how to create output files from input files. Dependencies between the rules are determined automatically". (Quote from Tutorial ) This means that snakemake needs to know which output file will be created by that rule. Then, it will determine whether it needs to run the rule. Therefore, the number of attempts should, at least usually, not be part of the output file name.

Maybe you want to combine the different files of the failed attempts? However, if the rule fails, then there will be no output file. Even if you force it. The file will be removed by snakemake. (See example below)

def getTargetFiles(files, attempted):
  return f"{files[:-4]}-{attempted}.txt"

rule combine:
  input:
    'test-1.txt'
  output:
    'test-combined.txt'
  shell:
    'cat test-[0-9]*.txt > test-combined.txt'

rule do_things_rule:
  resources:
    nr = lambda wildcards, attempt: attempt
  output:
    getTargetFiles("test.txt", 1)
  shell:
    'touch {output}; '
    'echo "Failing on purpose to produce file'
    '{output} at attempt {resources.nr}'
    '"; exit 1 '

How about leaving the number of attempts out of the filename and instead use resources.nr in the shell command?

Hope that this offers a solution to your problem.