Is it possible to cache a directory that is also an output?

[roaaaam]

Hi there,
I was able to try the cache a given directory that persist files between builds (or runs - as long as the name, worker, etc. do not change of course)
However when I also set the same directory as an output so that it's being used an the next task, the cache is no longer working.
Here is an example pipeline to reproduce:

---
jobs:
  - name: job-test-cache
    public: true
    plan:
      - task: write
        config:
          platform: linux
          image_resource:
            type: docker-image
            source: {repository: busybox}
          caches:
          - path: abc
          outputs:      # <- Cache do not work with this
          - name: abc   # <- Cache do not work with this
          run:
            path: sh
            args:
            - |
              ls
              echo 1 >> abc/xyz
              cat abc/xyz

      - task: read
        config:
          platform: linux
          image_resource:
            type: docker-image
            source: {repository: busybox}
          inputs:
          - path: abc
          run:
            path: sh
            args:
            - |
              ls
              cat abc/xyz

When I do not have the outputs and not the second task, I can see the xyz file growing on each run.

Is it possible to cache a directory that is meant to be outputed?

Thanks

[taylorsilva]

I don't think this would work. My guess as to what is happening under the hood is that Concourse creates two volumes:

• 1 volume for the cache you defined
• 1 volume for the output

Then when the container is made both these volumes get mounted at the same path inside your container. I also think this means the second volume to be mounted at that path is the volume that would get written to. I'm not sure if the order is deterministic or not, but based on what you've observed it sounds like it is.