Cache folder path doesn't exist on disk

[Bryan-Meier]

Description:
Something has changed in the last 2 weeks that is now causing the below error in the Post Setup Python action. For us, this is part of a deployment Github workflow and was working on the last deployment we did which was 2 weeks ago. And now is an issue on every deployment.

Error:
Cache folder path is retrieved for pip but doesn't exist on disk: /home/runner/.cache/pip. This likely indicates that there are no dependencies to cache. Consider removing the cache step if it is not needed.

Action version:
v5

Platform:

• [ X ] Ubuntu
• macOS
• Windows

Runner type:

• [ X ] Hosted
• Self-hosted

Tools version:
Python 3.11.17

Repro steps:
Run

Expected behavior:
No Post Setup Python error

Actual behavior:
Throwing this error:

Cache folder path is retrieved for pip but doesn't exist on disk: /home/runner/.cache/pip. This likely indicates that there are no dependencies to cache. Consider removing the cache step if it is not needed.

[priyagupta108]

Hello @Bryan-Meier 👋,
Thank you for your report. We'll investigate the issue and get back to you with the details!

[priyagupta108]

Hi @Bryan-Meier,

This is expected behavior and occurs when no dependencies are installed, so there’s no pip cache directory to save. This usually happens if your requirements.txt file is empty or if the install step is skipped, resulting in nothing to cache.

To address this:

• Make sure your workflow installs dependencies (e.g., with pip install -r requirements.txt).
• If your project doesn’t require dependencies, you can remove or conditionally skip the cache step. For example, simply omit the cache: pip line from your workflow.

Hope this helps! Let me know if you need further assistance.

[mturnerMOE]

Hi @priyagupta108,

We're seeing the error even with a filled requirements.txt and it's across multiple repos that haven't had any changes on our end. When looking through the logs, I can see a pip install -r requirements.txt being executed. Our jobs also run successfully as if dependencies are installed correctly.

[priyagupta108]

Hi @mturnerMOE,

Thanks for providing those details. It’s important that dependencies are installed successfully if you’re using caching, since the cache step depends on the pip cache directory being created during installation. If you’re certain that dependencies are installed (and you see pip install -r requirements.txt in the logs), but the error persists, there may be something unusual happening.

If you’re still facing this issue, could you please share a link to a workflow run where this error occurs? That will help us investigate further.

[Bryan-Meier]

Hi @priyagupta108 ,

Thanks for the response! For added context, @mturnerMOE and I work for the same company. I think I figured this out this morning. I'm still not sure why this has been working for 8 to 10 months and then all of a sudden started failing with zero changes on our side. I know there were no changes on our side because it effected every single repo we had and there are some repos that haven't been touched in about a year and some longer.

The issue seems to be that we have configured the setup-python@v5 step to use "cache: pip" and then when we install based on the requirements.txt, it's in the context of a virtual environment. I am assuming in that scenario cache is not used. I removed "cache: pip" and we no longer get the error. Below is what we are seeing.

Again, this had been working fine for all of our deployments with no change to our deployment workflow and then suddenly everything we deployed failed on the post step.

Thanks again for the response.

[simondeziel]

Same as @Bryan-Meier we see the "Post Setup Python" job fail from time to time and always when the pip cache dir was missing during the setup phase. https://github.com/canonical/lxd/actions/runs/16988588666/job/48162695778 is one such case.

It seems that if the setup phase does a cache miss and nothing causes the pip cache dir to be created, then the "Post Setup Python" will fail.

[priyagupta108]

Hi @Bryan-Meier,
Thanks for the clarification. I noticed you mentioned installing dependencies “in the context of a virtual environment,” but in the screenshot attached to your workflow, I don’t see a step that creates or activates a venv, and there’s also no pip install -r requirements.txt step shown.

To troubleshoot further, could you confirm if there’s a venv creation/activation and pip install -r requirements.txt step elsewhere in your full workflow? If those steps aren’t present, the cache: pip setting may not behave as expected.

Let us know if you can share the full workflow or clarify where the venv and pip install steps are happening!

Hi @simondeziel ,
We reviewed your workflow run and found that this issue occurs because documentation building (which includes running pip install and Sphinx setup) is intentionally skipped for pull request (PR) events in your script for the step Make LXD tarball and unpack it.

What’s happening:

• On PR events, make doc (and thus pip install) is not called, so no Python dependencies are installed and the ~/.cache/pip directory is never created.
• If the setup phase leads to a pip cache miss (such as when the cache was generated with Python 3.13.5 but the current runner uses Python 3.13.6, causing a cache key mismatch), and the pip cache directory hasn’t been created, the Post Setup Python step will fail when attempting to save the cache.
• For non-PR events (such as scheduled runs), the else block runs, pip install executes, the pip cache directory is created, and the cache step succeeds. See this scheduled run where pip install executes.
• In contrast, this PR run shows no pip install is run, so the pip cache directory does not exist.

Recommendation:
It’s important that dependencies are installed with pip during your workflow if you’re using caching, since the cache step depends on the pip cache directory being created. Make sure your workflow installs dependencies with pip install -r requirements.txt (or your method of choice) whenever caching is enabled.
For more details, see the setup-python advanced usage documentation on caching.

[simondeziel]

Hi @priyagupta108

Indeed, our workflow was sometimes mistakenly invoking the setup-python action and we've since fixed it. However, I think the setup-python action could more gracefully deal with ~/.cache/pip being missing and not fail but maybe log a warning indicating the cache was not used.

Thanks