Cache invalidated when $launchDir is used (no resume possible)

[tamasgal]

Bug report

I am not sure if this is a bug or feature, but it's definitely worth to mention, at least on https://www.nextflow.io/blog/2019/troubleshooting-nextflow-resume.html because I think that this behaviour is a bit unintuitive and took me quite a time to debug (I read through a lot of hash-trace diffs 🙈 )

When using the $launchDir meta variable (or $workflow.launchDir), the cache is somehow affected and not working when using -resume. I don't understand how this is achieved, but obviously nextflow knows that it's a path and treats it as an input. If the modification time of the folder has changed (which happens e.g. if a new file is created inside the folder), the cache is invalidated. Since nextflow writes its log-files by default in $launchDir/.nextflow.log*, no process can be resumed which is somehow using the $launchDir variable.

Expected behavior and actual behavior

Running the workflow multiple times with -resume does not use the cache when $launchDir is appearing in the script section. The $launchDir should not have any impact on the cache ingredients (hashes and timestamps of inputs/outputs).

Steps to reproduce the problem

Run the workflow below with nextflow run workflow.nf -resume multiple times, to see that the cache is not working.

#!/usr/bin/env nextflow
nextflow.enable.dsl = 2

process Process {
  input:
    val(input)
  output:
    file('*.txt')

  script:
      """
      echo $workflow.launchDir
      touch "a.txt"
      """
}

workflow {
  foo = Channel.from(1,2,3)
  Process(foo)
}

Program output

░ tgal@cca008:/sps/km3net/users/tgal/tmp/nextflow-cache
░ 18:38:25 > nextflow run workflow.nf -resume
N E X T F L O W  ~  version 21.10.3
Launching `workflow2.nf` [grave_waddington] - revision: 3ee6c5a013
executor >  local (3)
[49/67ffb1] process > Process (1) [100%] 3 of 3 ✔

░ tgal@cca008:/sps/km3net/users/tgal/tmp/nextflow-cache
░ 18:38:54 > nextflow run workflow.nf -resume
N E X T F L O W  ~  version 21.10.3
Launching `workflow2.nf` [kickass_volhard] - revision: 3ee6c5a013
executor >  local (3)
[f6/5b035c] process > Process (3) [100%] 3 of 3 ✔

Environment

• Nextflow version: 21.10.3.5655
• Java version: openjdk version "1.8.0_312", openjdk version "11.0.12" 2021-07-20 LTS
• Operating system: macOS Big Sur, CentOS 7, ArchLinux (2021.09)
• Bash version: zsh 5.0.2 (x86_64-redhat-linux-gnu), zsh 5.8 (x86_64-apple-darwin20.0)

[peddamat]

I have this same issue.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[pditommaso]

This happens because the launchDir is a directory Path. Every time the execution is launched, it will have different content and therefore it will produce a different hashing, causing the cache invalidation.

It you want to prevent that use workflow.launchDir.toString()