Hello,
I am seeking help in resolving intermittent connectivity issues to mirror.centos.org when running Jenkins CI jobs on CentOS Stream 9 (ppc64le) systems.
Context
• Our environment uses Conda, where we install R from source and build CRAN packages manually.
• On ppc64le, required *-devel packages (like gdal-devel, proj-devel, etc.) are not available in RHEL repos, so we depend on CentOS Stream 9 repos for these.
• On x86, Anaconda provides prebuilt R and CRAN packages, so these RPM dependencies are not needed.
Problem
We’re encountering random connection refused errors when accessing mirror.centos.org during dnf operations in Jenkins CI pipelines.
Could not connect to mirror.centos.org: Connection refused
• These failures are not persistent — simply restarting the job resolves the issue.
• Retrying the failed step inside the pipeline does not help, as the issue continues during that session.
Root Cause
• mirror.centos.org uses GeoIP-based redirection to assign a mirror.
• Once a redirect is made, the selected mirror is pinned for the duration of the session.
• If that mirror is temporarily unhealthy or down, dnf fails with a connection error.
• Jenkins jobs reuse the same mirror during retries within the pipeline, so the failure persists.
• Only restarting the entire Jenkins job triggers a new redirect, which often resolves the issue. Looking For Help With
• Is this a known issue affecting CentOS mirrors for ppc64le?
• Are there recommended static mirrors we can switch to for better reliability?
• Are there best practices for using CentOS repos in automated pipelines on Power architecture?
Looking For Help With
• Is this a known issue affecting CentOS mirrors for ppc64le?
• Are there recommended static mirrors we can switch to for better reliability?
• Are there best practices for using CentOS repos in automated pipelines on Power architecture?