Clear working directory between runs?

[TaxBusby]

Hi there, I'm using the non-ephemeral runners, which is fine with me for performance/isolation tradeoff.

However, it looks like the working directory given to workflows does not get cleared between jobs. This can lead to surprise for developers who expect a clean working directory.

Is there an option available to have the runners clear the working directory between runs?

---

Here's a short workflow I'm using to test this case:

name: Ephemerality Test
on:
  workflow_dispatch:

jobs:
  write:
    runs-on: [self-hosted]
    name: Write
    steps:
      - name: Write files
        run: |
          echo Current Directory: $(pwd)
          ls
          echo "Local File Exists" > hello.txt
          echo "Temp File Exists" > /tmp/hello.txt
          
  read:
    runs-on: [self-hosted]
    needs: [write]
    name: Read
    steps:
      - name: Test for persisted files
        run: |
          echo Current Directory: $(pwd)
          ls
          cat hello.txt || echo "Local dir cleared"
          cat /tmp/hello.txt || echo "Tmp dir cleared"
          
          if [ -f hello.txt ]; then
              echo "Local file was not cleared!"
              exit 1
          fi
          if [ -f /tmp/hello.txt ]; then
              echo "Temp file was not cleared!"
              exit 1
          fi

Outputs:

// Write
Current Directory: /opt/actions-runner/_work/gh-self-hosted-runners/gh-self-hosted-runners
hello.txt

// Read
Current Directory: /opt/actions-runner/_work/gh-self-hosted-runners/gh-self-hosted-runners
hello.txt
Local File Exists
Temp File Exists
Local file was not cleared!

[sykhro]

This is the standard behaviour when using self-hosted runners with GitHub Actions. You are responsible for cleanup in your own workflows

[winwinashwin]

From their blog post, github seems to recommend using pre/post job hooks for cleanup. Currently the terraform does not allow the user to override or extend the pre job hook created by the terraform.

1. The pre job script is overridden and not appended to.
2. The script does not execute any additional user configurable scripts.
3. The start-runner.sh is executed after both pre and post user-data scripts which the user can provide as inputs to the terraform.

All these make it really difficult to extend the pre job hook.

@npalm Can we add a feature where the user can extend or replace the pre job script?