[BUG] CI instability due to external Gin service outages (HTTP 503)

[nishanthcr7777]

Describe the bug

CI tests intermittently fail due to transient HTTP 503 errors when accessing atlas files hosted on gin.g-node.org. These failures are not caused by recent code changes and do not reproduce locally.

To Reproduce

1. Open a pull request that triggers GitHub Actions CI.
2. Run workflows on ubuntu-latest.
3. Observe intermittent failures in tests that rely on Gin-hosted atlas files.

Expected behaviour

CI should pass reliably. Temporary outages of external services should not cause unrelated pull requests to fail.

Actual behaviour

CI occasionally fails with:

• HTTP 503 Service Unavailable during atlas downloads
• PermissionError when interacting with read-only last_versions.conf

Failing tests

• test_get_download_size_no_size_url
• test_get_download_size
• test_conf_from_url_read_only

Log file
CI logs show:

• requests.exceptions.HTTPError: 503 Server Error
• PermissionError: [Errno 13] Permission denied: last_versions.conf

Additional context

• All tests pass locally
• Re-running CI often resolves the failure
• Failures correlate with Gin service availability

Potential improvement
Consider adding test-level safeguards such as:

• Retry/backoff or HEAD-only checks for network calls
• Marking network-dependent tests as xfail/skip when services are unavailable
  This would improve CI reliability without affecting production behavior.

[adamltyson]

Hi @nishanthcr7777 we're aware of this, and are moving data storage to S3 to improve the issue.

I don't think we should xfail or skip these tests though, as we won't notice when they fail for other reasons. I will keep the issue open though so that others can see that we have a plan :)

[nishanthcr7777]

Thanks for the update @adamltyson .

Agreed on not skipping or xfailing the tests. I’m happy to leave things as it is on my side.

Thanks for keeping the issue open and for the clarity!